release: prepare deterministic 1.1.0 silence detection

[BTawaifi]

Summary

• keep the known-good fixed -38 dBFS FFmpeg detector as the default path
• add opt-in deterministic --threshold auto analysis with conservative peak windows and exit hysteresis
• add optional --hysteresis_db for numeric thresholds
• make stereo/multichannel detection conservative: activity on any channel protects the interval
• fall back to FFmpeg filter-script files for very large edit graphs / command-line length safety
• expose resolved threshold + detector in ProcessingResult
• bump package version to 1.1.0
• add CHANGELOG and expand README with FFmpeg install, tuning, _silent.wav inspection-artifact warning, multichannel and large-file behavior

Safety / compatibility

• no model, VAD, training data, internet service, or cloud dependency
• default numeric threshold path is unchanged
• retained audio rendering remains timing-only (atrim / timestamp reset / concat)
• no normalization, limiting, EQ, fades, crossfades, compression, or gain changes
• auto mode never chooses a threshold more aggressive than the stable -38 dBFS default

Validation

• local regression suite: 24/24 passing
• fixed default path on the supplied WhatsApp Opus fixture produces the exact same decoded non-silent WAV bytes as the current main output
• auto mode uses peak rather than RMS windows so short loud consonant/transient samples break silence instead of being averaged away