feat(vid1): support multi-track audio in VID1 GameCube video/audio

[bebnkarnate]

Summary

VID1 movie containers (THAW GameCube) can carry multiple dubbed-language audio tracks — one AUDH header per track in HEAD, and one AUDD block per track per FRAM. The parser previously only read the first AUDH header, then merged every AUDD chunk in each frame into a single packet list, corrupting audio on any multi-track file (English/French/Italian/German/Spanish tracks interleaved together into one broken stream).

This is directly relevant to interview/cutscene clips that ship with multiple language dubs, such as the movies/vid interview files in Tony Hawk's Downhill Jam (Wii).

Changes

• Vid1AudioExtractor now parses every AUDH track in HEAD and correctly routes each frame's AUDD blocks to their matching track by position, instead of merging them.
• Added ProbeTracks / per-track Probe / ConvertToWav overloads for track-level access. ConvertToWav converts every track by default, writing one WAV per track ({stem}.wav, {stem}_track1.wav, …).
• Vid1VideoConverter now muxes every audio track into the exported MP4 as its own selectable AAC stream (previously only the first track was muxed in).

Video preview fix

While verifying the multi-track MP4 export, found that exported MP4s failed to preview both externally (Discord's embed transcoder) and inside this app's own GUI (MediaPlayerElement). Root cause: raw RGB24 frames were piped straight into libx264 with no output pixel format specified, so ffmpeg auto-selected the High 4:4:4 Predictive profile (yuv444p) to match the RGB input. That decodes fine in VLC/ffplay-class software decoders, but not in restrictive decoders like Discord's transcoder or Windows Media Foundation (which the GUI preview relies on) — both expect standard 4:2:0 chroma subsampling.

Fixed by forcing -pix_fmt yuv420p (standard chroma subsampling) and -movflags +faststart (moves moov to the front so playback/preview can start without downloading the whole file) in both native-decode ffmpeg pipelines: Vid1VideoConverter (VID1 GameCube) and StrConverter (PS1 STR), which shared the same raw-RGB-into-libx264 pattern. Verified via ffprobe: profile went from High 4:4:4 Predictive/yuv444p to standard High/yuv420p, with moov now ordered before mdat.

Verification

Tested against a real multi-track sample (JX_Interview01.vid, Tony Hawk's Downhill Jam Wii):

• Parses 5 audio tracks correctly (matching the expected 5-language dub set).
• Produces 5 correctly separated WAV files via Vid1AudioExtractor.ConvertToWav.
• Produces an MP4 with 5 distinct AAC audio streams via Vid1VideoConverter.ConvertToMp4 (verified with ffprobe).

Added synthetic multi-track test coverage (Vid1TestBuilder.CreateMultiTrackVid1) plus real-sample tests that run when the sample file is present, and skip gracefully otherwise (tests/NeversoftMultitool.Tests/Core/Formats/Audio/Vid1AudioExtractorTests.cs).

Full test suite: no new regressions. Two pre-existing failures remain (Parse_AllThawGcSamples_ClassifiesLongFormAndAtvi, TryWriteDeterministicVideoStream_AllThawVidSamples_WritesNonEmptyM4v) — both expect all 8 real .vid samples in movies/vid/, but this environment only has 1; unrelated to this change.