Add pacing, backoff, duplicate detection, and structured logging to recognition loop

[juaneloti]

Summary:

• Added structured logging (INFO + DEBUG) for each recognition attempt (match / no_match / error) with sleep timing.
• Implemented pacing: hold period after a successful match (default 5 min), minimum interval between unknown attempts, jitter to reduce bot‑like patterns.
• Added exponential backoff and longer pause after repeated failures; prevents hammering Shazam endpoints.
• Skips duplicate audio by hashing recorded snippet (SHA1) to avoid re-sending identical data too soon.
• Hardened audio capture: falls back to silence if device errors/overflows, preventing thread crash.
• Resilient config loader now tolerates inline # comments and missing keys (adds pacing defaults).
• Added safe cover art fetch with timeout and logging; history writing wrapped in try/except.

Why:

Improves stability, reduces risk of rate limiting, makes troubleshooting easier, and avoids wasted recognition calls.