Task: Implement endpoints to support series page

[LiteralGenie]

• Link the series folders on disk with metadata from the db
• Endpoint that exposes this info

[LiteralGenie]

I'd like to keep the user info (the series on disk) separate from the MU data, but if instead I store the raw MU data separately (ie json, not normalized into tables), I can make this a future problem.

[LiteralGenie]

So to link series on disk with the MU metadata, the series (folder) name has to be compared with the titles in the MU data.

This implies some kind of similarity metric. As it turns out, the fuzzyset package will be the fastest for calculating this (see table below or string_comp_results.md for details).

At ~10ms per query, this implies ~100s for 10k series. With multiprocessing, this further implies ~25s with 4 workers and ~8s with 12 workers.

	queries.json	5	20	100
jellyfish-damerau_levenshtein	673	285	676	2717
jellyfish-hamming	88	83	85	88
jellyfish-jaro	317	133	305	897
jellyfish-jaro_winkler	315	135	306	894
jellyfish-levenshtein	349	141	354	1739
textdistance-damerau_levenshtein	510	833	493	430
textdistance-hamming	1441	453	353	452
textdistance-jaro	495	510	382	506
textdistance-jaro_winkler	448	456	451	385
textdistance-levenshtein	445	429	445	447
fuzzyset	11	2	7	34
sqlite3-editdist3	2446	894	2508	10400

[LiteralGenie]

Scanning disk with Path.glob() is also pretty slow. Turns out Path.iterdir() is much faster because(?) it doesn't fetch metadata like file size.

(eg scanning 20k folders takes 15.3s vs 0.8s, and this is after running a few times so that whatever disk cache kicks in.)