feat(audio): add --ntsc_audio_rate for NTSC-locked PCM output

[mpownby]

Full disclosure: Claude did all the work; I reviewed code changes, they looked minimal/safe. We (Claude and me!) did manual testing to confirm that the .pcm output file size changes as expected when this flag is passed.

Add an opt-in --ntsc_audio_rate flag that outputs analog audio locked to NTSC line timing (2.8 samples/line = 1470 samples/frame, ~44055.944 Hz) instead of the default 44100 Hz, which drifts at 1471.47 samples/frame against NTSC video.

This reuses the existing negative-freq convention in downscale_audio (a negative value is a multiple of the horizontal line frequency). The multiple 2.8 is fractional, so two plumbing fixes were needed:

• core.py: stop truncating analog_audio with int() for negative values, so the fractional HSYNC multiple survives.
• core.py: resolve the negative rate to its real nominal value in build_json so metadata reports the true sample rate (44055.944...) rather than the -2.8 sentinel.

PAL is already frame-locked at 44100 Hz (1764 samples/frame), so the flag is NTSC-only and warns + no-ops on PAL. Default behavior is unchanged for backward compatibility.

Add tests/test_audio_rate.py covering the negative-freq mapping, and document the new flag in docs/user-guide/command.md.

Checklist

• I have searched the open pull requests to confirm this change has not already been submitted.
• My branch is up to date with the target branch.
• I have tested my changes and all existing tests pass.
• I have updated documentation where necessary.
• My code follows the project's coding standards (see CONTRIBUTING.md).

Description

Adds an opt-in --ntsc_audio_rate flag that outputs analog audio locked to NTSC
line timing — exactly 2.8 samples/line (1470 samples/frame, 735/field), i.e.
~44055.944 Hz — instead of the default 44100 Hz. Off by default; existing
behavior is unchanged.

Motivation

Primary motivation is to improve disc audio stacking.

The default 44100 Hz PCM rate does not divide evenly into NTSC frame timing:
44100 / (30000/1001 fps) ≈ 1471.47 samples/frame, so the audio sample grid
slowly drifts relative to the video. 44055.944 Hz = 2.8 × the NTSC horizontal
line frequency (4.5 MHz / 286 = 15734.2657 Hz), giving exactly 1470 samples per
frame and perfect, drift-free A/V alignment. PAL already divides evenly
(44100 / 25 fps = 1764 samples/frame), so it needs no change.

Related Issues

N/A

Changes Made

• Add --ntsc_audio_rate CLI flag (default off, backward compatible). NTSC-only;
  warns and no-ops on PAL.
• Reuse the existing negative-freq convention in downscale_audio (a negative
  value is a multiple of the horizontal line frequency) by mapping the flag to
  freq = -2.8.
• core.py: preserve fractional negative rates (no longer int()-truncate
  analog_audio), so -2.8 survives end-to-end.
• core.py build_json: report the resolved nominal sample rate
  (44055.944055944055) in the .tbc.json instead of the -2.8 sentinel.
• Add tests/test_audio_rate.py and document the flag in
  docs/user-guide/command.md.

Testing

• All existing tests pass (pytest) — 5 passed.
• Tested manually with: a CAV NTSC capture (-s 2000 -l 30). Default →
  44144 stereo samples (1471.47/frame); --ntsc_audio_rate → exactly 44100
  stereo samples (1470.000/frame). JSON sampleRate reported 44100 vs
  44055.944055944055.
• New tests added for: the negative-freq → samples-per-line mapping
  (-2.8 → 1470/frame, -2.8 → 44055.944 Hz, default 44100 → 1471/frame).

Screenshots (if applicable)

N/A — no UI changes.

Additional Notes

• The .tbc.json sampleRate can now be a non-integer float when this flag is
  used. Downstream ld-decode-tools should accept a float sampleRate (handled
  separately).
• PAL output is intentionally unaffected.