Validation: reference startup music against emulator or hardware capture

[akratch]

Problem\nNative music is now playable and much improved, but final fidelity still needs reference-driven validation rather than only human listening. Startup music is the best broad sample because it covers a long boot-to-start-menu sequence and exposes the remaining thin/bright/reverb differences.\n\n## Desired end state\n- A contributor can capture the boot-to-start-menu audio from a trusted emulator or hardware setup without committing any copyrighted audio.\n- GE007_MUSIC_AUDIO_DUMP output from MGB64 can be compared against that local reference with tools/compare_audio_reference.py.\n- The documented comparison reports useful segmented band/envelope metrics and identifies the remaining mixer/ABI stages to investigate.\n\n## Suggested lane\nUse docs/RELEASE_CHECKLIST.md and docs/INSTRUMENTATION.md as the starting point. Keep all WAV/raw captures under /tmp or another ignored local path.\n\n## Validation\n- python3 tools/compare_audio_reference.py /tmp/ge007_reference.wav /tmp/ge007_music.raw --test-format raw --segment-seconds 10 --segment-hop-seconds 5 --print-bands --print-segments\n- ./scripts/ci/check_release_ready.sh\n\n## Asset rule\nDo not upload or commit reference audio, ROM-derived music dumps, or videos containing game audio.