Tourmaline: analysis moves to web worker

Analysis pipeline moved from server-side computation to a web worker. The previous approach POSTed scrobbles to /api/profile/[did], but Vercel's 4.5MB serverless function body limit meant users with 50K+ scrobbles (~8.5MB JSON) got silently empty profiles.

The web worker runs the full pipeline (Aggregator, genre/mood/era/timeline profiling, diversity, obscurity, sessions, personality, story recap) off the main thread. Scrobbles pass from client memory to the worker via structured clone — no network round-trip, no body size limits. The worker bundle is 56KB. Client strips scrobbles to essential fields (playedTime, trackName, releaseName, artists[0].name, duration, musicServiceBaseDomain) before posting to the worker.

Two dead ends preceded this: first, a GET endpoint that relied on server-side SQLite cache (ephemeral on Vercel — each serverless invocation gets a fresh filesystem), then a POST endpoint that hit the body limit. The worker is the right architecture — the data is already in the browser, the computation is pure, and there are no deployment constraints.