Find a lesson that teaches a given grammar concept.

Matches the concept against LessonPassage.key_grammar[].point. The two naming sources are NOT normalized — lesson points are verbose/free-form ("Simple Present Tense for facts and routines"), sometimes multilingual dicts, while the progress breakdown uses a different, terser inventory ("Present Simple"). So we don't substring-match; we score TOKEN OVERLAP after normalizing both (lowercase, strip parentheticals, drop stopwords + over-generic grammar words like "tense"/"verb"). Returns the passage id to open at /lessons/{id}.