Fix #57: drop dead _app- filter and refresh hardcoded /api/bleed fallbacks

[dpashutskii]

Summary

Fixes #57.

HLTB rotated the search endpoint again (/api/finder → /api/bleed). Runtime discovery in SearchInformations adapts to whatever endpoint name shows up in the JS, but the existing _app-*.js script-name filter on the first pass of send_website_request_getcode (and its async sibling) hasn't matched anything since HLTB moved to Turbopack — chunks now have opaque names like 0-~-0up.q3_p0.js. So every search was failing the first pass, falling back to the second pass, and either succeeding slowly or hitting the stale /api/s hardcoded fallback.

What changes

1. Drop the _app-*.js script filter in send_website_request_getcode and async_send_website_request_getcode. The methods now iterate every <script src> tag and stop at the first one that yields a search_url. The parse_all_scripts parameter is preserved for backward compatibility (no longer changes behaviour, documented in docstring).
2. Refresh hardcoded fallbacks: SEARCH_URL and SearchAuthToken.search_url updated from /api/s to /api/bleed (current as of 2026-05-07).
3. Bug fix in the async path: previously a non-200 response from any single script caused return None to break out of the loop entirely. Now skips and continues to the next script, matching the sync version's intent.
4. New unit tests in tests/test_search_url_extraction.py — 5 hermetic tests covering the discovery regex against the real /api/bleed chunk shape, a hypothetical future rename, GET-only fetches, scripts with no fetch call, and versioned endpoints (/api/bleed/v2). These don't hit the network so they're safe to run in CI without flakes.
5. Version bump 1.0.21 → 1.0.22.

The integration tests in tests/test_normal_request*.py are unchanged.

Verification

$ python3 -m unittest tests.test_search_url_extraction -v
test_extracts_a_hypothetical_future_endpoint ... ok
test_extracts_current_api_bleed_endpoint ... ok
test_extracts_root_path_from_versioned_endpoint ... ok
test_ignores_get_only_fetches ... ok
test_returns_none_when_no_post_fetch_present ... ok
Ran 5 tests in 0.000s
OK

End-to-end against live HLTB:

>>> from howlongtobeatpy import HowLongToBeat
>>> results = HowLongToBeat().search("The Witcher 3")
>>> max(results, key=lambda r: r.similarity).game_name
'The Witcher 3: Wild Hunt'
>>> max(results, key=lambda r: r.similarity).main_story
51.61

Why minimal-diff

I kept the parse_all_scripts parameter on both *_getcode methods rather than removing it, so any downstream caller that was passing True/False directly continues to work. Same goes for the duplicate retry call in send_web_request / send_async_web_request — it's now wasteful but harmless, and removing it would expand the diff. Happy to drop those if you'd prefer a cleaner cut.

🤖 Generated with Claude Code

[ScrappyCocco]

I appreciate the support a lot but I would like you to clean the code a little
If you don't have the time I can do it as soon as I have the time

Additionally you should add your test as ignored by codecov.yml (I think it's the only place in which the test file is referenced)

[ScrappyCocco]

This was intentionally /s so that if the "retrieve url" doesn't work it's easily noticeable because this variable was not changed

[ScrappyCocco]

Same reason as above + remove the comments
I think both here and above no change is required

[ScrappyCocco]

This is a nice change but it inverted the "if" condition for no reason, redo the condition as before please so the diff is more noticeable

[ScrappyCocco]

Also are you sure it's not working? See comment in the issue #57 (comment) - maybe I don't need these changes