Add a persistent index of user uploads for incremental ThePosterDB scrapes

[Thoroslives]

Scraping a ThePosterDB user currently re-fetches their entire uploads listing every run. For a user with thousands of uploads that is hundreds of page requests just to rediscover posters we have already seen. This adds an opt-in cache_user_scrapes option that keeps a small SQLite index (standard library sqlite3, one file in the config directory) of each scraped user's uploads. User pages are newest-first, so a repeat scrape only fetches pages until it reaches a page consisting entirely of already-indexed uploads, records anything new, then processes the user's full catalogue from the index. The results (filters, excludes, labels, log lines and counts) are the same as a full crawl, it just skips re-downloading pages it has already seen. Steady-state re-runs of a big user drop from hundreds of page requests to two, which is also much kinder to ThePosterDB.

To stay honest against edits, deletions and ordering surprises, every user_cache_refresh_days days (default 7) the next scrape of that user quietly re-crawls every page and reconciles the index (uploads that have disappeared are tombstoned and no longer applied). The scrape also cross-checks ThePosterDB's own upload counter against the index each run and falls back to a full re-crawl on any unexplained difference. --no-cache forces a full crawl for one run (CLI, bulk lines and the web UI), and the index is stored with a normalized title key so future features can look cached artwork up by title.

• cache_user_scrapes in config.json (default false, so behaviour is unchanged unless enabled) plus a toggle in the web UI; user_cache_refresh_days (default 7)
• Applies to ThePosterDB user scrapes only; set/poster URLs and MediUX are untouched
• No new dependencies; a few pytest unit tests for the new index logic under tests/ (using the markers already defined in pytest.ini)
• README updated

[Thoroslives]

Pushed a couple of follow-up commits to this branch:

• Stop the cached crawl at the end of the uploads. With the persistent index, a user whose TPDB upload counter overshoots the pages it actually lists was crawling on into empty pages every run; the parse error there marked the crawl "unclean", which blocked both the crawl-state record and the tombstone pass, so the affected user did a full crawl every time and never reconciled. Now the crawl stops at the first page that fetches cleanly but lists nothing.
• Guard the tombstone pass. A full crawl only reconciles (tombstones assets it didn't see) when it actually covered most of the user's reported uploads; a crawl cut short by a bad page records a non-full crawl and retries next run instead of deleting what it didn't reach.
• Report new cache entries in the run summary.

Note: the "stop at an empty page" part relies on get_posters() no longer raising on a page with no poster grid, which I've submitted separately as #61 (it's a general crawl bug independent of the index). If you're happy with both, merging #61 first keeps this one a clean rebase.

[Thoroslives]

Rebased onto v0.8.8 for clean mergeability.