Score against original user intent on Reddit, remove the artificial low-end relevance floor, and make Polymarket semantics dominate generic market quality signals.
Also apply the relevance filter to Polymarket and update the affected cross-source tests.
Validation: uv run python -m unittest
Consolidate duplicated _extract_core_subject() (7 copies across bird_x,
reddit, youtube_yt, tiktok, instagram, bluesky, scrapecreators_x) into
query.extract_core_subject() with parameterized noise set, max_words,
and suffix stripping.
Consolidate duplicated _tokenize/_compute_relevance/STOPWORDS/SYNONYMS
(4 copies across youtube_yt, tiktok, instagram, scrapecreators_x) into
relevance.token_overlap_relevance() with hashtag-aware matching.
Integration into per-module imports follows in next commits.
Claire Novotny