fix: improve max amplitude detection and global_gain range

[M-Igashi]

Summary

Fixes #51 and partially addresses #52.

Changes

Issue #51: find_max_amplitude reports incorrect values

• Fixed iterate_frames to skip Xing/Info VBR header frames

  • These frames contain metadata, not audio data
  • Their global_gain values (often 0) were incorrectly included in range calculation
  • Now matches mp3gain behavior

• Added find_peak_amplitude function in replaygain.rs

  • Decodes audio using Symphonia to measure actual PCM peak values
  • Returns peak in normalized range (0.0-1.0+) and PCM scale (0-32768+)

• Updated find_max_amplitude to use audio decoding

  • When replaygain feature is enabled, decodes audio for accurate peak measurement
  • Falls back to global_gain estimation when feature is disabled

• Output format now shows Max PCM sample in 32768 scale

  • Compatible with mp3gain output format

Issue #52: MP3 peak value capped at 1.0

• Root cause identified: Symphonia's MP3 decoder internally clamps output samples to [-1.0, 1.0] in synthesis.rs
• Workaround implemented: Added warning for MP3 files when peak is at clipping threshold
  • Informs users that actual peak may be higher than reported
  • AAC/M4A files correctly report peaks > 1.0 (no warning needed)

Testing

Tested with various MP3 and M4A files:

# MP3 file (clipping warning shown)
$ mp3rgain -x file.mp3
  Max PCM sample: 32768.000000
    (may be clipped - actual peak could be higher)
  Max global_gain: 182
  Min global_gain: 110

# M4A file (no warning, correct peak > 32768)
$ mp3rgain -x file.m4a
  Max PCM sample: 41531.960938
  Headroom: -2.06 dB

Limitations

Issue #52 cannot be fully resolved without modifying Symphonia's MP3 decoder. The decoder's clamp at [-1.0, 1.0] is a design decision in symphonia-bundle-mp3/src/synthesis.rs:

*o = s.clamp(-1.0, 1.0);

Potential future solutions:

1. Submit a PR to Symphonia to add an option for unclamped output
2. Use a different MP3 decoder that doesn't clamp