log: later discs and series parts survive the log reader

[cursor]

Bug and impact

Two log-mode reader bugs left after #15 / #18:

1. Later discs skipped. Each fix.csv row is processed on its own, but the never-expiring __cache__/book marker was keyed only on the book column. After grouping discs under the release folder, a later run (or a name-only marker from a prior hybrid pass) printed Skipping: … already processed for every remaining disc. cd2 never reached the library.

2. Series part lost. Matched Audible/MAM rows still used setSeries(seriesparts), which splits on #. The log writes seriesparts as Name part. A matched row with a blanked paths column (the documented re-pin workflow) filed as Series 17.5 - Title and wrote that name into metadata.opf.

Root cause

• isCached / cacheMe hashed MAMBook.name only. Log mode creates one MAMBook per row (hash("{i}-{book}")) but they share the same name.
• #18 taught setSeriesFromLog to pair series + seriesparts for the id3 columns; the audible/mam rebuild in buildTreeFromLog was not switched over.

Fix

• Log-mode rows set cacheKey to {book}|{fullpath}; getHashKey uses that when present. Hybrid mode is unchanged (still hashes the release name).
• Audible/MAM series reconstruction uses setSeriesFromLog like the id3 columns.

Validation

• Two-phase unit test: cd1-only run, then both rows — cd2 is filed; the already-filed disc is skipped. Confirmed the same test fails under the old name-only key.
• Already-matched Pike Logan 17.5 row with blank paths files as Pike Logan #17.5 - Title and the OPF has calibre:series / series_index.
• Full suite: 223 tests OK. Correctness and security review of the diff: clean after the test was rewritten to the incremental trigger.