feat(sampler): play bitmap video at the source fps on the Ultimate Audio sampler

[kfox]

What

Follow-up to the Ultimate Audio sampler integration (#46): re-tune the default frame rate for bitmap (hires/mhires) video whose audio runs through the FPGA PCM sampler, plus a small audio write-ahead-lead hardening.

1. Source-rate playback for sampler video

The sampler plays off the C64 bus and its presence forces the bus-clean REU-staged (bank-swap) video path. So sampler-audio bitmap video needs neither:

• the 4-bit $D418 DAC's 20 fps cap (which exists only because the DAC's NMI + ring writes compete with frame uploads for the bus), nor
• the muted half-rate (30/25) cap (host-DMA tear avoidance).

_frame_push_default_fps gains an off_bus_audio path that uncaps to the system rate (60 NTSC / 50 PAL) — but as a poll ceiling. Because VideoScene dedups (it re-pushes only when the source yields a new frame), the effective push rate equals the source video's own fps:

source	effective push
24 fps clip	24/s
30 fps clip	30/s
60 fps clip	60/s

i.e. faithful source-rate playback capped at the VIC refresh — no artificial cap, no dropped frames, no wasted re-pushes. DAC video stays at 20; muted bitmap stays at 30/25; the non-sampler paths are untouched.

2. Lead-headroom decouple (sampler audio)

Decouple the sampler's startup prebuffer (stays 0.5 s → prompt start) from its runtime write-ahead lead target (0.5 s → 1.0 s). The lead is buffer depth, not A/V latency (video tracks the audio read head), so a deeper target rides out heavier PyAV decode stalls without shifting sync.

3. Diag harness

scripts/diags/run_and_capture.py gains an --app-arg passthrough (e.g. --app-arg=-v to surface the sampler write-ahead-lead telemetry) and tees the app log; HDMI captures now downscale to ~960 px before write.

Verification

• make check green (ruff + mypy --strict + pyright + 1664 unit tests, incl. new off_bus_audio helper + build_scene wiring cases and prebuffer/lead-decouple cases).
• Verified on real U64 hardware via HDMI capture: real ≤30 fps content pushes at its source rate with no added shimmer, audio stays clean at a genuine 60/s push, and the lead floor on a heavy 4K clip roughly doubled with the decouple.

Notes / follow-up

• Continuous-motion shimmer scales with push rate and appears only on true >30 fps sources — that's the separate, pre-existing unsynced-bank-swap-timing issue, not this fps default.
• A distinct A/V-sync-on-heavy-clips investigation is queued as a follow-up: the lag is supply-side (full-resolution 4K decode can't produce frames in real time), to be measured via software instrumentation (audio-clock vs displayed-frame PTS) — orthogonal to this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.52%. Comparing base (ac0c58e) to head (0d567b7).
⚠️ Report is 2 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #48      +/-   ##
==========================================
+ Coverage   79.49%   79.52%   +0.03%     
==========================================
  Files          69       69              
  Lines       13491    13496       +5     
  Branches     2013     2014       +1     
==========================================
+ Hits        10724    10733       +9     
+ Misses       2303     2300       -3     
+ Partials      464      463       -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.