Rework BookWalker and Kodansha providers

[ThoughtzThruKeyz]

Ports the BookWalker and Kodansha provider reworks from loocool2/komf, source changes only.

Based on merge/upstream-prs (PR #2), not master. Merge that one first, or retarget this to master and expect the earlier commits to come along.

Why

Both providers are broken against the live services — verified, not assumed:

Provider	Old endpoint	Status now
Kodansha	api.kodansha.us	No DNS records at all. Fully dead.
BookWalker	global.bookwalker.jp	301 → bookwalker.com/migration/, which serves a Cloudflare JS challenge ("Just a moment…"). Server-side scraping cannot pass it.

Neither was deprecated by upstream; both broke because the external services changed.

What changed

Kodansha — searches via the site's WordPress endpoint (kodansha.us/wp-json/kodansha/v1/search-series, verified 200 JSON) and parses series/volume detail from server-rendered HTML via a new KodanshaHtmlParser. This trades a JSON API for scraping, which is inherently more fragile — but the API is gone, so it's scraping or nothing. Also fixes volume synopsis fetching.

BookWalker — replaces scraping with BookWalker's own published catalog export at static.bookwalker.com/data/bkwk-db.sqlite.zst. Volumes now render as "title, Vol. x".

Verified against the live file:

• HTTP 200, application/zstd, 52,871,942 bytes, Last-Modified today, ETag present
• zstd magic 28 b5 2f fd; decompressed prefix is SQLite format 3
• Every table the new code queries exists in the real export (contributors, labels, product_contributors, product_external_ids, products, publishers, series, series_tags, tags). series_fts is the trigram index the downloader builds itself.

Download behavior: conditional on ETag/Last-Modified, written to generation-named files so in-flight readers aren't disturbed, refreshed in the background on a 24h interval with failures non-fatal.

Things to know before running this

• Disk and bandwidth. ~52 MB download, ~233 MB extracted, plus ~14 MB index. New bookwalkerDatabaseDir config option, default ./bookwalker.
• Cold start blocks. With nothing on disk, the first BookWalker query waits on the full download and decompress. Subsequent staleness refreshes are backgrounded.
• commons-io is used but declared nowhere — it resolves transitively through commons-compress. It packages fine today (443 classes in the shadowJar), but it's an undeclared dependency worth pinning later.

Deliberately excluded

• Their build files — that fork sits 3 commits behind upstream and re-adds jvmToolchain(17), which upstream removed in 8bcdea0 and your fork already has.
• Their Dockerfile (Java 25, drops arm32), release workflow, and README.
• Their MangaBaka changes, to keep this scoped. Worth revisiting: the titles v2 field is real (16 variants with language/is_primary) and would improve name matching. Their extensionless-cover fix is harmless but its premise didn't reproduce — the URL returns an identical 200 with and without .jpg.
• Everything from PR Add Kavita 0.9.x+ compatibility fixes Snd-R/komf#309 that PR Merge upstream PRs #322, #316, and the code portion of #309 #2 already excluded. That fork merged Add Kavita 0.9.x+ compatibility fixes Snd-R/komf#309 wholesale and still carries the .wasm blobs, the Google Drive extension link, the removed COEP/COOP headers, and a submodule pointer (bf519dcf) that exists in no public Komelia repo — so git submodule update --init cannot succeed there.

Test plan

• :komf-core, :komf-mediaserver, :komf-app compile clean
• :komf-app:shadowJar assembles; zstd-jni natives and commons-io both packaged
• Remote export verified as zstd SQLite with the expected schema
• Confirmed no MangaBaka, Dockerfile, workflow, README, or submodule drift in the diff
• Runtime: full download + first BookWalker match against a real library
• Runtime: Kodansha search/match against a real library