Low-hanging improvements to the crawler APIs

[jodal]

Low-hanging improvements to the crawler APIs, found by reading LxmlParser,
FeedParser and all 60 hand-written crawl() implementations. One commit per
change, so they can be reviewed or dropped individually.

Dead code and wrong types in LxmlParser

• DoesNotExist was raised nowhere, so the handler wrapping _get_all() could
  never run. Both are gone.
• _decode() could never take its bytes branch, as lxml returns str from both
  Element.get() and text_content(). That it was applied in _get_all() but
  not _get_one() showed nothing depended on it.
• text() was annotated as returning list[str] | str | None, while its
  overloads and _get_one() only ever produce str | None.

Errors that were classified wrong

• The three shared base crawlers asserted the parser had found an image URL,
  preempting the ImageURLNotFound that add_image() raises. That turned a
  CrawlerBroken naming the comic and date into a bare AssertionError in the
  catch-all, and would vanish under python -O.
• LxmlParser fetched pages without checking the status, so an error page was
  parsed as if it were the comic's page. Finding no image in it, a crawler
  reported "no release found" — which is what a day the comic did not publish on
  looks like — so a site that moved or started refusing us stayed invisible.
• Three crawlers that already establish the entry is the comic, by its link,
  tags or publishing date, skipped to the next entry when the image selector
  then found nothing. They now let the empty URL reach CrawlerImage, which
  reports ImageURLNotFound at error level.

Two additions that shrink the crawlers

• first=True on the singular accessors takes the first match in document order
  instead of raising MultipleElementsReturned. Pages that legitimately match
  several had to fall back to the plural accessor and index into it, guarding the
  empty list by hand. Seven crawlers move over.

• element() / elements() return parsers scoped to the matching elements, for
  crawlers that need to pick a container and then read its children. Four
  crawlers come off page.root, where they hand-rolled the extraction and lost
  the default handling, the multiple-match guard and the CSS selectors. This also
  fixes a latent bug in the Evil Inc crawler, which tested xpath() against
  None while it returns an empty list.

  Awkward Zombie and Subnormality keep using page.root: they match on element
  text and sort by a parsed style attribute, which the selector API does not
  express.

Smaller things

• Entry.title and Entry.link reach crawlers through __getattr__, so they
  typed as Any and nothing checked their use in the 41 feed crawlers. Declared
  alongside the existing summary and content0 annotations.
• The OOTS crawler called feed.all() twice, rebuilding the entry list to test
  for and then take the first entry.
• history_length_days = 0 resolves to the same history_start as leaving both
  history attributes out, which already means only today can be crawled.

Worth watching after deploy

raise_for_status() is the one change that can turn a working crawler into a
failing one, if a site serves usable content with a non-2xx status. I checked the
crawlers that mention HTTP status: Penny Arcade's soft-404 comes with a 200 and
is unaffected, Dumbing of Age's 404 note is about the image server on the
downloader path, and Buttersafe's User-Agent dodges a 403 that would now surface
as a warning instead of being parsed silently.

Not touched, but the same shape

• Wumo's guard is now safe to drop too, since a gone page raises rather than
  falling through.
• CreatorsCrawlerBase also calls httpx.get() without a status check and then
  .json(), so a non-2xx fails as a JSON decode error and lands in the catch-all
  rather than as a CrawlerHTTPError.

LxmlParser and FeedParser have no tests, so the new first=True and
elements() surface ships unit-tested only by hand against sample markup.

https://claude.ai/code/session_013DLpoXp8z36AGE2ksWG58n

[codecov]

Codecov Report

❌ Patch coverage is 36.11111% with 23 lines in your changes missing coverage. Please review.
✅ Project coverage is 69.00%. Comparing base (81b4208) to head (e1c8f66).
⚠️ Report is 12 commits behind head on main.

Files with missing lines	Patch %	Lines
src/comics/aggregator/lxmlparser.py	32.35%	23 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #698      +/-   ##
==========================================
+ Coverage   68.97%   69.00%   +0.02%     
==========================================
  Files          46       46              
  Lines        2069     2071       +2     
==========================================
+ Hits         1427     1429       +2     
  Misses        642      642              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.