feat: add ja.rawdevart

[kinboy56]

fix #34

Summary

• Search, browse/filtering, series details, chapter list, and page list
• 3 listings: Latest, Popular, Trending Today
• Filters: sort (3 options), status, and genre (fetched dynamically, 60+ options)
• Deep linking for both series and chapter URLs

Implementation notes

A few site quirks that aren't obvious from the API responses and are worth a reviewer's
attention:

• The /spa endpoints require no special headers. The site's own scripts send a Referer
  and X-Requested-With, but a direct request without either returns a byte-identical response
  (verified by diffing both). Plain Request::get is used throughout; only the image requests
  set a Referer, since the CDN mirrors do require one.
• Chapter pages carry no image urls in the api's HTML fragment as-is — the markup uses
  <img class="lazy" data-src="..."> with paths relative to a server field the same response
  provides (plus slaves as mirrors). abs: attribute resolution isn't used to join them,
  since it depends on the parsed document keeping a base url, which a Html::parse_fragment
  fragment doesn't reliably carry; the join is done by hand instead.
• Chapters are keyed by chapter_number, not chapter_id. The page endpoint is
  /spa/manga/{manga_id}/{chapter_number}; passing a chapter_id there returns a null
  chapter_detail (confirmed against the live api). Chapter numbers are non-integer for roughly
  85% of entries in the sampled catalogue (e.g. 34.2) and were confirmed to be unique and
  non-null across every chapter checked, so they're safe to use as the key.
• Genres are fetched instead of hardcoded. DynamicFilters reads the <option> markup the
  /spa/genre/all response embeds under genreOpt, so new genres the site adds don't need a
  source update. The "all" default plus every id past it are validated as numeric in a test.
• manga_status can be entirely absent from the API response, not just true/false.
  Missing status maps to MangaStatus::Unknown rather than defaulting to "ongoing".
• Chapter language is left unset on purpose. Setting it (as the previous, now-removed
  version of this source did) subjects chapters to the app's chapter-language filter, which can
  hide every chapter depending on the user's device settings for a single-language source. Most
  existing ja.* sources in this repo leave it unset for the same reason.
• Series and chapter urls prefix the numeric id with two letters (e.g. /g/ne854721) that
  the site's routing ignores; the api only ever wants the digits. DeepLinkHandler strips the
  prefix and rejects anything that isn't left with a purely numeric id, and also trims query
  strings / fragments first, since shared links commonly carry tracking parameters.
• Content rating isn't provided by the api. It's inferred from genre tags
  (adult/hentai/loli-family tags → NSFW, ecchi → Suggestive, otherwise Safe), since the
  source itself is registered as contains-nsfw.

Test plan

• cargo test --release — 9 tests, all hitting the live api (no mocked responses), covering:
  all 3 listings, search, filter combinations, dynamic genre filter parsing, series details
  (title/cover/authors/tags/description/status), a chapter with a non-integer number round-tripped
  into a page request, page list urls, deep link parsing (including tracking-parameter and
  non-numeric-id variants), and a regression test against the specific series/chapter pair
  originally reported as unreadable
• cargo fmt --check — clean
• cargo clippy --release — no warnings
• aidoku package + aidoku verify — schema validation passes for source.json and
  filters.json, icon is 128×128 and fully opaque
• [x ] Manual on-device install via aidoku serve — not yet done in this environment

Checklist

• The source can be compiled without any warnings or errors
• cargo fmt has been run before submission
• cargo clippy outputs no lint warnings
• All files have an additional newline at the end
• JSON files use tabs for indentation
• status / content_rating are set on series when available (status is Unknown when
  the api doesn't provide it, rather than guessed)
• DeepLinkHandler is implemented for both series and chapter urls
• Genre filtering uses the site's own options via DynamicFilters, kept in sync automatically
• Home page / listing components beyond simple listings — not implemented (optional per
  CONTRIBUTING.md)