v2.9.2

Highlights

Fix: ReplayGain no longer over-attenuates sparse material (#217). Silent 50 ms windows were being dropped from the loudness histogram instead of counted, which shrank the 95th-percentile denominator and made tracks with significant silence — acapellas, vocal stems, sparse mixes — read up to ~1.5 dB too loud and recommend too much attenuation. Silent windows are now clamped and counted, matching the original gain_analysis.c. Dense music is unaffected. If you analyzed acapellas or stems with an earlier version, re-run the analysis.

Other Changes

• Add an isolated cross-check test verifying mp3rgain's ReplayGain analysis matches the reference gain_analysis.c to floating-point precision, independent of the decoder (#201)
• deps: bump anyhow from 1.0.102 to 1.0.103 (#215)
• ci: bump the github-actions group with 2 updates (#216)

Thanks to @dgilman for the verification push (#201) that surfaced #217.

Full Changelog: v2.9.1...v2.9.2