fix(physics): replace fitted flipper transfer

Remove the live-fitted upward and downward flipper polynomials. Port the
reconstructed 1000:7ed9 path with delta-specific pivots and record edges,
integer cross-product and radius gates, penetration, response-record gain,
wrapping velocity updates, and the final position delta.

Use resting records on press and the already-raised records on release, correct
the original delta direction, and apply each edge to both live ball slots. Keep
render-facing Vec2 projections, but adjust their float representation by ULPs so
every gameplay millipixel round-trips exactly instead of losing reconstructed
integer results.

Test Plan:
- `bash original/tools/test_reconstructed_c.sh` -- passed during raised-record reference capture
- `cargo test --all-targets` -- passed, 62 tests
- `cargo clippy --all-targets -- -D warnings` -- passed
- `rumdl check tdkpin-rs/CHANGELOG.md tdkpin-rs/RECONSTRUCTION.md` -- passed
- `git diff --cached --check` -- passed

tdkpin-rs/RECONSTRUCTION.md

@@ -22,7 +22,7 @@ implementation.
 | Artwork | Exact | All 34 custom DIB images, three standard bitmaps, icon, and palette derivatives are preserved in `assets/original/`. The game uses the original 640x460 table, loading, help, ball, wheel, robot, plunger, media, and diamond frames. |
 | Audio | Exact samples and recovered dispatch | All 16 mono PCM WAV resources are embedded unchanged. Playback is emitted at the reconstructed call sites, including multi-sound bank completions and nudge-then-tilt ordering. WAV 2022 is loaded by the original generic resource loop but has no playback call and is therefore never played by Rust. |
 | Help and languages | Exact | Original resource images 1001-1005 are displayed directly. |
-| Playfield collision layout | Recovered | All 109 active type-2 line objects and 40 static active type-1 circles are transcribed from the original 175-object registration table. The registration routine converts its sideways inputs with `screen = (y, x - 20)` and accumulates explicitly relative objects. Type-2 records retain every recovered Real48 normal/tangent response pair and registered one-sided orientation. Type-1 records retain their swept-circle radius, radial rebound, tangent coupling, and bumper kick. Each flipper uses its exact two line records plus moving tip circle in both positions. The upward edge transfer matches three live probes and the downstroke matches its live position-preserving `(3519,6302)` transfer; both are mirrored on the right. Object 174 is omitted because the original overwrites it with the live ball every frame. |
+| Playfield collision layout | Recovered | All 109 active type-2 line objects and 40 static active type-1 circles are transcribed from the original 175-object registration table. The registration routine converts its sideways inputs with `screen = (y, x - 20)` and accumulates explicitly relative objects. Type-2 records retain every recovered Real48 normal/tangent response pair and registered one-sided orientation. Type-1 records retain their swept-circle radius, radial rebound, tangent coupling, and bumper kick. Each flipper uses its exact two line records plus moving tip circle in both positions. Moving-flipper contact ports `1000:7ed9` rather than fitting live samples: delta-specific pivots/edges, integer cross gates, radial/penetration calculations, response-record gain, and position/velocity publication are tested against all four C harness directions and the raised release geometry. Object 174 is overwritten with the live first ball and Rust handles its ball-to-ball role directly. |
 | Ball launcher and nudge input | Recovered | The initial 32-bit fixed-point coordinates decode to `(325, 413)` in the right shooter lane. Each Down keydown subtracts `15*50 = 750` millipixels, release subtracts another `15*100 = 1500`, and the result follows the recovered randomized `-3800` lower and `-2280` weak upper clamp branches. The ten decoration frames use the same strict 750-millipixel thresholds. Left Shift and keypad 3 apply their directional `(50-Random(20))*15` impulses; Space uses the recovered Real48 horizontal factor and `(60-Random(20))*15` vertical impulse. Each nudge adds 25 to the wrapping 16-bit tilt counter, compares it with `30+Random(10)`, and the detail timer decrements a nonzero counter once per callback. |
 | Physics arithmetic | Recovered gameplay behavior | Production movement uses the original 10 ms millipixel substep, `+15` vertical acceleration, `3800` speed bound, point-path type-2 intersection, one-sided line response, swept type-1 circle response, and swept non-physical sensor contacts. It evaluates all records and applies the earliest contact along the substep. The original Borland seed update and high-word `Random(n)` mapping drive launcher variation, effects, claw terminals, and the recovered randomized magnetic-field impulse. Radius-bearing type-4 triggers independently scale X and Y by `1.03-Random*0.08` after their event path. Live probes cover ordinary rails, ordinary circles, a kicked bumper, lock holes, magnetic fields, all claw exits, and both flipper directions. Rust represents the original per-ball contact words as path-entry/inside latches; this is a source-structure difference rather than a missing collision route. |
 | Rules | Recovered gameplay paths | Player count, controls, the five three-line bumper-value groups, four three-line TDK-diamond groups, five doubling-value lock holes, wheel-reset target, seven-way effect selector/consumer including multiball, permanent double scoring, and four exact media/extra-ball thresholds follow original help/code paths, globals, and object flags. The ninth diamond pays the original 24,464 completion value; the following completed bank enables double scoring, and later completions add 100,000 to the per-player secondary score. Turn changes mirror the original save/load of all 175 collision record states: wheel/top targets, active/contact slots, selected effect, and multiball readiness remain attached to their player. Claw contact and all initially active type-4 targets use recovered records. The top three targets score 500 each and independently enable the left, center, or right magnetic field record; each field pulls the ball upward until it exits and then deactivates. The claw state machine and release table have live differential coverage for all four random terminals. Remaining timing uncertainty is presentation batching at non-default detail settings, not gameplay routing. |
@@ -72,10 +72,9 @@ decoded, build-ready subset; it does not replace that evidence archive.
   covers repeated launches, both flippers, targets, bumpers, lock holes, a claw
   capture/release pair, and drains while checking every state for finite values.
 - Semantic boundary: gameplay routes and initialized values are covered, but no
-  claim is made that every host floating conversion or repaint tick is binary-
-  identical to Borland Real48 and Win16 GDI execution. Live-fitted flipper
-  transfer and non-default repaint batching remain documented source/platform
-  differences.
+  claim is made that every remaining host `f64` intermediate or repaint tick is
+  binary-identical to Borland Real48 and Win16 GDI execution. Non-default repaint
+  batching remains a documented platform difference.
 
 ## Architecture