fix(metadata): restore author/series catalog cache on SQLite

[kevinheneveld]

Summary

The persisted author and series metadata cache never returns a hit on SQLite, so every author and series page re-fetches the full catalog from Audible on each load. The cache read orders by a Nullable<DateTime>.GetValueOrDefault(...) expression the SQLite EF Core provider can't translate; the query throws, the caller's best-effort catch swallows it as a cache miss, and the cache is effectively dead. This switches the ordering to a translatable COALESCE and adds regression coverage that runs against the real SQLite provider.

Why this matters

Open any author or series page twice. The catalog should come from the persisted cache the second time, but instead each visit hits Audible again — slower pages, and avoidable load on the upstream Audible endpoints (and more exposure to their rate limits). On a library with many authors/series, ordinary browsing keeps re-fetching catalogs that were already cached minutes ago.

The cache looks like it works in tests because the EF InMemory provider happily evaluates the same LINQ that SQLite rejects — so the failure only appears against a real database.

Implementation notes

• Surface: listenarr.infrastructure/Persistence/Repositories/AudiobookRepository.cs — four cache-read methods (GetCachedAuthorByNameAsync, GetCachedAuthorByAsinAsync, GetCachedSeriesByNameAsync, GetCachedSeriesByAsinAsync).
• Change: OrderByDescending(e => e.LastFetchedAt.GetValueOrDefault(e.UpdatedAt)) → OrderByDescending(e => e.LastFetchedAt ?? e.UpdatedAt). The null-coalescing form maps to SQL COALESCE(LastFetchedAt, UpdatedAt); identical "freshest-first" ordering semantics, but translatable. An inline comment at each site records why, so the pattern doesn't regress.
• No behavior change beyond the cache now actually being read: same ordering, same rows, same return shape.
• Testing approach: a new test class deliberately uses the real SQLite provider (Microsoft.Data.Sqlite :memory:) rather than the EF InMemory provider, because InMemory can't reproduce the translation failure. The new tests fail against the current code with Translation of method 'System.DateTime?.GetValueOrDefault' failed and pass with the fix.

What this PR does NOT include

• No changes to the cache write paths or the cache schema.
• No unrelated catalog/series UI features — this is the bug fix in isolation.

Test plan

• dotnet test tests/Listenarr.Tests.csproj --filter FullyQualifiedName~AudiobookRepository_CatalogCacheReadTests — 3/3 passing (new).
• Verified the new tests fail on the current canary with the GetValueOrDefault translation error — they guard the actual regression.
• dotnet test tests/Listenarr.Tests.csproj — full suite 685/685 passing.
• Frontend untouched.

[therobbiedavis]

@kevinheneveld This looks good to me. Please delete the changelog and update the PR description to match the template.