fix: make smart defaults zero-config

[BTawaifi]

Summary

• route the installed audio-desilencer entry point through a dedicated CLI module
• make CLI --threshold auto the default while preserving the processor/library fixed-threshold behavior for backward compatibility
• keep speech-oriented defaults at 700 ms minimum silence, 150 ms retained pause, 3 dB auto hysteresis, WAV output
• add regression coverage for zero-config CLI defaults and fixed-threshold overrides
• update README and changelog so audio-desilencer input.m4a is the recommended normal usage

Why

The previous package entry point still exposed the older parser behavior on installed systems, which made --threshold auto fail when an old build was installed and also required unnecessary tuning for normal speech recordings.

Compatibility

• Python API behavior remains available with explicit thresholds
• --threshold -38 still uses the stable fixed FFmpeg path
• no ML/model/training/cloud dependency is added