epignosis: verify Audnexus enrichment for audiobooks

[forkwright]

Context

Audiobook m4b files sometimes have sparse iTunes-style metadata (no narrator, no series, no year). Audnexus is the canonical metadata source used by AudioBookshelf and the broader audiobook self-hosting community. `epignosis` is supposed to use Audnexus for audiobook enrichment.

The reference Python migration script at `~/menos-ops/scratch/migrate-audiobooks-canonical.py` deferred Audnexus enrichment with a TODO and falls back to filename parsing — it flags affected books with `needs_metadata = true` in `audiobook.toml`.

What

1. Verify the Audnexus provider is wired in `epignosis` for audiobook enrichment
2. Add tests for the lookup paths:
   • Lookup by ASIN (when available)
   • Lookup by title + author (when ASIN missing)
3. Wire it into the audiobook import path so that when m4b atoms are sparse (missing year, narrator, series), `epignosis::resolve_identity` and `epignosis::enrich` reach out to Audnexus
4. Cache Audnexus responses per-provider per the existing epignosis caching pattern

Files

• `crates/epignosis/src/providers/` — verify or add Audnexus provider
• Cache layer
• Tests

Acceptance

• Audnexus provider exists and is registered for audiobook media type
• Lookup-by-ASIN works
• Lookup-by-title+author works
• Cached responses
• Audiobook import path uses Audnexus when m4b atoms are sparse
• Test against a real audiobook with known ASIN (e.g. Brandon Sanderson's The Way of Kings, ASIN B003P2WO5E)

Why this matters

Once landed, the `needs_metadata = true` flag in legacy migrated audiobook sidecars can be swept by re-running `harmonia migrate audiobooks` (issue: `harmonia migrate subcommand`) and the Audnexus calls will populate the missing fields automatically.

Reference

• Audnexus API: https://api.audnex.us/
• Reference fallback implementation: `~/menos-ops/scratch/migrate-audiobooks-canonical.py` (filename-only fallback with TODO comment for Audnexus)