perf(catalog): parallelize production manifest hashing

Production catalog generation previously prepared every game serially and computed transfer-chunk hashes for extracted .eti files even though those hashes were discarded. Enable Rayon-backed BLAKE3 hashing, process selected games through the shared available-CPU pool, and hash large read blocks with update_rayon. Extracted outputs now use a whole-file-only hash path. Manifest collection remains deterministic and the independent second verification pass is preserved.

Test Plan:
- `just clippy` -- passed
- `just test` -- passed

Cargo.toml

@@ -13,7 +13,7 @@ members = [
 
 [workspace.dependencies]
 base64 = "0.23"
-blake3 = "1"
+blake3 = { version = "1", features = ["rayon"] }
 bytes = { version = "1", features = ["serde"] }
 cap-fs-ext = { version = "4", default-features = false }
 cap-primitives = "4"
@@ -25,6 +25,7 @@ if-addrs = "0.15"
 log = "0.4"
 mdns-sd = "0.20"
 mimalloc = { version = "0.1", features = ["secure"] }
+rayon = "1"
 rcgen = {
   version = "=0.14.8",
   default-features = false,