feat: Use sl_source field to improve BookDB trust weighting

[deucebucket]

Problem

When BookDB returns audio identification results, it includes an sl_source field indicating the quality of the match:

• database — matched against 50M+ book database (highest confidence)
• audio — Whisper transcript only, no database match (medium)
• live_scrape — last-resort web scraping (lowest)

LM currently captures this field but never uses it. All three sources get treated identically in the BookProfile, missing an opportunity to improve confidence scoring.

Fix

In library_manager/providers/bookdb.py and library_manager/pipeline/layer_audio_id.py:

• Map sl_source to differentiated source names: bookdb_audio (database), bookdb_transcript (audio), bookdb_scrape (live_scrape)
• Use appropriate weights: database=75, transcript=60, scrape=45
• Store the differentiated source in BookProfile so confidence calculations reflect actual match quality

Files

• library_manager/providers/bookdb.py — map sl_source to differentiated source names
• library_manager/pipeline/layer_audio_id.py — pass differentiated source to BookProfile
• library_manager/models/book_profile.py — add new source weights if needed