feat(compress): split MT stream drain policy

The MT stream initializer previously exposed one C callback that waited for
all submitted jobs and then released their resources. That left the ordering
policy in C even though Rust already owned the surrounding initialization
sequence and the ring-order wait/release policies.

Expose separate wait-all and release-all callbacks at the Rust/C boundary and
invoke them in Rust in the original wait-before-release order. C retains the
worker synchronization, job descriptors, input buffer, and resource-pool
side effects behind the two callbacks, while the focused Rust initializer test
now makes the ordering explicit.

Test Plan:
- capped cargo +nightly fmt for the Rust workspace -- passed
- capped root cargo clippy --all-targets -- -D warnings -- passed
- capped serial make -j1 for the single-threaded library, MT library, and CLI -- passed
- git diff --cached --check -- passed

lib/compress/zstdmt_compress.c

@@ -756,7 +756,8 @@ typedef struct {
     unsigned allJobsCompleted;
 } ZSTDMT_RustInitCStreamProjection;
 typedef size_t (*ZSTDMT_initResizeFn)(void* opaque, unsigned nbWorkers);
-typedef void (*ZSTDMT_initDrainFn)(void* opaque);
+typedef void (*ZSTDMT_initWaitForAllJobsFn)(void* opaque);
+typedef void (*ZSTDMT_initReleaseAllJobResourcesFn)(void* opaque);
 typedef void (*ZSTDMT_initApplyParametersFn)(void* opaque, size_t jobSize);
 typedef size_t (*ZSTDMT_initDictionaryFn)(void* opaque);
 typedef void (*ZSTDMT_initSetSizeFn)(void* opaque, size_t size);
@@ -938,7 +939,9 @@ void* ZSTDMT_rust_createCCtx(
         ZSTDMT_createCCtxFreeFn freeContext);
 size_t ZSTDMT_rust_initCStream(
         const ZSTDMT_RustInitCStreamProjection* projection, void* opaque,
-        ZSTDMT_initResizeFn resize, ZSTDMT_initDrainFn drain,
+        ZSTDMT_initResizeFn resize,
+        ZSTDMT_initWaitForAllJobsFn waitForAllJobs,
+        ZSTDMT_initReleaseAllJobResourcesFn releaseAllJobResources,
         ZSTDMT_initApplyParametersFn applyParameters,
         ZSTDMT_initDictionaryFn prepareDictionary,
         ZSTDMT_initSetSizeFn setTargetPrefixSize,
@@ -2782,13 +2785,16 @@ static size_t ZSTDMT_initCStreamResize(void* opaque, unsigned nbWorkers)
     return ZSTDMT_resize(state->mtctx, nbWorkers);
 }
 
-static void ZSTDMT_initCStreamDrain(void* opaque)
+static void ZSTDMT_initCStreamWaitForAllJobs(void* opaque)
 {
     ZSTDMT_initCStreamState* const state = (ZSTDMT_initCStreamState*)opaque;
-    ZSTDMT_CCtx* const mtctx = state->mtctx;
+    ZSTDMT_waitForAllJobsCompleted(state->mtctx);
-    ZSTDMT_waitForAllJobsCompleted(mtctx);
+}
-    ZSTDMT_releaseAllJobResources(mtctx);
+
-    mtctx->allJobsCompleted = 1;
+static void ZSTDMT_initCStreamReleaseAllJobResources(void* opaque)
+{
+    ZSTDMT_initCStreamState* const state = (ZSTDMT_initCStreamState*)opaque;
+    ZSTDMT_releaseAllJobResources(state->mtctx);
 }
 
 static void ZSTDMT_initCStreamApplyParameters(void* opaque, size_t jobSize)
@@ -3031,7 +3037,8 @@ size_t ZSTDMT_initCStream_internal(
     return ZSTDMT_rust_initCStream(
         &projection, &state,
         ZSTDMT_initCStreamResize,
-        ZSTDMT_initCStreamDrain,
+        ZSTDMT_initCStreamWaitForAllJobs,
+        ZSTDMT_initCStreamReleaseAllJobResources,
         ZSTDMT_initCStreamApplyParameters,
         ZSTDMT_initCStreamPrepareDictionary,
         ZSTDMT_initCStreamSetTargetPrefixSize,

rust/src/zstdmt_compress.rs

@@ -1166,7 +1166,8 @@ pub struct ZSTDMT_initCStreamProjection {
 }
 
 pub type ZSTDMT_initResizeFn = unsafe extern "C" fn(*mut c_void, c_uint) -> usize;
-pub type ZSTDMT_initDrainFn = unsafe extern "C" fn(*mut c_void);
+pub type ZSTDMT_initWaitForAllJobsFn = unsafe extern "C" fn(*mut c_void);
+pub type ZSTDMT_initReleaseAllJobResourcesFn = unsafe extern "C" fn(*mut c_void);
 pub type ZSTDMT_initApplyParametersFn = unsafe extern "C" fn(*mut c_void, usize);
 pub type ZSTDMT_initDictionaryFn = unsafe extern "C" fn(*mut c_void) -> usize;
 pub type ZSTDMT_initSetSizeFn = unsafe extern "C" fn(*mut c_void, usize);
@@ -2766,12 +2767,14 @@ fn init_rsync_parameters(target_section_size: usize) -> (u64, u64) {
 /// Run the high-level MT streaming initialization policy while C retains all
 /// private state and side effects behind callbacks.  The callback order is
 /// intentionally the order of the original C initializer: resize, normalize,
-/// drain, attach the first dictionary, size buffers, reset stream state,
+/// wait for all jobs, release their resources, attach the first dictionary,
+/// size buffers, reset stream state,
 /// update the active dictionary, and finally reset serial state.
 #[inline]
 fn init_c_stream_with<
     Resize,
-    Drain,
+    WaitForAllJobs,
+    ReleaseAllJobResources,
     ApplyParams,
     PrepareDictionary,
     SetPrefixSize,
@@ -2785,7 +2788,8 @@ fn init_c_stream_with<
 >(
     projection: ZSTDMT_initCStreamProjection,
     mut resize: Resize,
-    mut drain: Drain,
+    mut wait_for_all_jobs: WaitForAllJobs,
+    mut release_all_job_resources: ReleaseAllJobResources,
     mut apply_parameters: ApplyParams,
     mut prepare_dictionary: PrepareDictionary,
     mut set_target_prefix_size: SetPrefixSize,
@@ -2799,7 +2803,8 @@ fn init_c_stream_with<
 ) -> usize
 where
     Resize: FnMut(c_uint) -> usize,
-    Drain: FnMut(),
+    WaitForAllJobs: FnMut(),
+    ReleaseAllJobResources: FnMut(),
     ApplyParams: FnMut(usize),
     PrepareDictionary: FnMut() -> usize,
     SetPrefixSize: FnMut(usize),
@@ -2825,7 +2830,8 @@ where
     );
 
     if projection.allJobsCompleted == 0 {
-        drain();
+        wait_for_all_jobs();
+        release_all_job_resources();
     }
 
     apply_parameters(job_size);
@@ -3181,7 +3187,8 @@ pub unsafe extern "C" fn ZSTDMT_rust_initCStream(
     projection: *const ZSTDMT_initCStreamProjection,
     opaque: *mut c_void,
     resize: Option<ZSTDMT_initResizeFn>,
-    drain: Option<ZSTDMT_initDrainFn>,
+    waitForAllJobs: Option<ZSTDMT_initWaitForAllJobsFn>,
+    releaseAllJobResources: Option<ZSTDMT_initReleaseAllJobResourcesFn>,
     applyParameters: Option<ZSTDMT_initApplyParametersFn>,
     prepareDictionary: Option<ZSTDMT_initDictionaryFn>,
     setTargetPrefixSize: Option<ZSTDMT_initSetSizeFn>,
@@ -3198,7 +3205,8 @@ pub unsafe extern "C" fn ZSTDMT_rust_initCStream(
     };
     let (
         Some(resize),
-        Some(drain),
+        Some(wait_for_all_jobs),
+        Some(release_all_job_resources),
         Some(apply_parameters),
         Some(prepare_dictionary),
         Some(set_target_prefix_size),
@@ -3211,7 +3219,8 @@ pub unsafe extern "C" fn ZSTDMT_rust_initCStream(
         Some(serial_reset),
     ) = (
         resize,
-        drain,
+        waitForAllJobs,
+        releaseAllJobResources,
         applyParameters,
         prepareDictionary,
         setTargetPrefixSize,
@@ -3230,7 +3239,8 @@ pub unsafe extern "C" fn ZSTDMT_rust_initCStream(
     init_c_stream_with(
         projection,
         |nb_workers| unsafe { resize(opaque, nb_workers) },
-        || unsafe { drain(opaque) },
+        || unsafe { wait_for_all_jobs(opaque) },
+        || unsafe { release_all_job_resources(opaque) },
         |job_size| unsafe { apply_parameters(opaque, job_size) },
         || unsafe { prepare_dictionary(opaque) },
         |size| unsafe { set_target_prefix_size(opaque, size) },
@@ -7768,7 +7778,8 @@ mod tests {
         let events = Rc::new(RefCell::new(Vec::<&'static str>::new()));
         let values = Rc::new(RefCell::new(Vec::<(&'static str, usize)>::new()));
         let resize_events = Rc::clone(&events);
-        let drain_events = Rc::clone(&events);
+        let wait_events = Rc::clone(&events);
+        let release_events = Rc::clone(&events);
         let apply_events = Rc::clone(&events);
         let apply_values = Rc::clone(&values);
         let prepare_events = Rc::clone(&events);
@@ -7793,7 +7804,8 @@ mod tests {
                 resize_events.borrow_mut().push("resize");
                 0
             },
-            move || drain_events.borrow_mut().push("drain"),
+            move || wait_events.borrow_mut().push("wait"),
+            move || release_events.borrow_mut().push("release"),
             move |job_size| {
                 apply_events.borrow_mut().push("apply");
                 apply_values.borrow_mut().push(("job", job_size));
@@ -7837,7 +7849,8 @@ mod tests {
             events.borrow().as_slice(),
             &[
                 "resize",
-                "drain",
+                "wait",
+                "release",
                 "apply",
                 "prepare-dict",
                 "prefix",
@@ -7865,6 +7878,7 @@ mod tests {
             init_projection(),
             |_| expected_error,
             || panic!("a failed resize must stop initialization"),
+            || panic!("a failed resize must stop initialization"),
             |_| panic!("a failed resize must stop initialization"),
             || panic!("a failed resize must stop initialization"),
             |_| panic!("a failed resize must stop initialization"),
@@ -7891,6 +7905,7 @@ mod tests {
             projection,
             |_| panic!("workers already match"),
             || panic!("all jobs are complete"),
+            || panic!("all jobs are complete"),
             {
                 let events = Rc::clone(&events);
                 move |_| events.borrow_mut().push("apply")