Add command-scoped index cache and cancellable HTTP requests

[polymorcodeus]

Summary

SQLite-derived search index moves from a package-level variable into a command-scoped indexCache, and the web scraping path gains explicit context.Context support with a 10-second timeout. This makes the read and network paths explicit, cancellable, and safe for concurrent test runs.

What's new

• Command-scoped index cache - cmd/book/index.go introduces indexCache, created in Main and closed by a deferred cache.close(). All index-using commands receive *indexCache explicitly.
• Explicit cache plumbing - shelves, collections, marks, searchMarks, gc, doctor, runIndexRebuild, runIndexSync, and indexStaleFiles now take the cache as a parameter instead of reaching for global state.
• Context-aware HTTP requests - web.WebsiteTitle and web.LoadWebsite accept a context.Context, and WebsiteTitle builds an http.Client capped at 10 seconds so cancellation and timeouts are always enforced.
• Propagated command context - book mark add passes its action context into web.LoadWebsite, replacing the previous implicit context.Background() call site.

Behavior changes / guardrails

• Main owns the single index database connection for the process lifetime; no other code path opens and closes its own connection.
• The previous package-level index *catalog.Index variable is removed entirely.
• Concurrent go test runs no longer share mutable index state.
• The 10-second effective timeout for title fetching is preserved; only cancellation behavior is added.

Testing

• Updated internal/web/web_test.go to pass context.Background() into WebsiteTitle.
• Updated cmd/book/actions_test.go to pass a context into addMark.
• make check runs fmt, vet, lint, and tests; all green.