fix(diarization): upgrade bootstrap to pyannote.audio>=4 (fixes #3)

[AlexanderAbramovPav]

Summary

Fixes #3 — diarization failed on fresh installs with TypeError: Pipeline.from_pretrained() got an unexpected keyword argument 'token' (and 'plda' after patching to use_auth_token).

Root cause

whisply 0.14.1 hard-pins pyannote.audio==3.4.0 / whisperx==3.7.8. The bootstrap only ran uv pip install "whisply[mlx]", so fresh installs landed on pyannote 3.4.0. But the diarization path loads pyannote/speaker-diarization-community-1 with token=, which require pyannote.audio ≥ 4.0 (3.x uses use_auth_token and lacks community-1's PLDA clustering). My venv was manually on the tested 4.0.4 / 3.8.6 baseline, so it never surfaced locally.

Fix

• transcribe.sh: bootstrap now upgrades to whisperx>=3.8.6 + pyannote.audio>=4.0, and self-heals venvs created before this fix (fast idempotent major-version probe).
• transcribe_whisperx.py: preflight version guard → actionable message + graceful skip on pyannote < 4, instead of a cryptic crash. import whisperx made lazy so the module is unit-testable.
• Tests: cover the version guard (runs in the pytest+numpy-only CI).
• Docs: setup.md install + troubleshooting; CHANGELOG; plus a docs/adr-faster-whisper.md noting faster-whisper is already the whisperX engine (direct integration rejected).

Verification

• pytest tests/ → 51 passed (48 existing + 3 new).
• uv pip install --dry-run --upgrade "whisperx>=3.8.6" "pyannote.audio>=4.0" resolves cleanly over whisply's == pins (no whisperx/pyannote downgrade).