[BACKEND] Scoring + feed endpoints — interest/popularity/timeliness ranking with no AI

[linear]

Depends on Phase 2 (classifier + event_tags). Delivers a complete, shippable personalized feed with no AI — keyword ranking only.

Build steps

1. Write server/lib/scoring.ts — combines interest match, popularity (save/rsvp/view/click counts), timeliness, and featured boost; apply a seen penalty per user (optional, see notes)
2. Implement feed endpoints:
   • GET /feed/home — Upcoming section + one carousel per bucket the user has selected
   • GET /feed/explore — All events ranked by popularity + timeliness
   • GET /feed/bucket/:id — Events for a single bucket, personalized rank
3. Wire Home and Explore screens in the app to the new endpoints
4. Delete stale client-side taxonomy logic in home.tsx

Notes

• This is the early-launch cut line — a complete feed ships here with no AI dependency
• Seen penalty is weak signal until scroll-level impressions are tracked; treat as optional for now
• Test: logged-in user sees personalized carousels, ranked correctly

[linear]

LOOP-222