Fix: ship /api/plans/:id/activity endpoint

The bug: a437 (2026-08-03) promised this endpoint but the route was never built. SPA called it from 3 places (list mount-time chips, detail activity card, detail linked activity) and all 3 silently 404'd. Every plan card showed … loading forever on the state chip.

The fix (src/mavis-hq.js): shipped the endpoint with the 4-state machine from a437:

• `pre_coded` — no ticks yet
• `in_progress` — at least one step is status=in_progress
• `done` — meta.status=completed OR all steps done
• `paused` — no in_progress step AND no non-keep-alive heartbeat in 15+ min

The keep_alive carve-out: VPS-timer pings every 60s on every live thread. Without filtering, no plan would ever look paused. The endpoint now:

• queries `recentHqHeartbeats(thread_id, 50)`
• shows the most recent 5 (with keep_alive flag in response)
• computes last_hb_at from the FIRST non-keep_alive row only
• pause override fires on the real heartbeat gap

Response shape:

{ ok, plan_id, computed_state, state_explanation,
  current_step, owner, recent_heartbeats,
  last_finished_bulletin, paused }

Also fixed: SPA mount-time fetch for state chip ran BEFORE c.innerHTML was set, so chips didn't exist in DOM. Moved the forEach to after the innerHTML assignment.

Verification (Playwright iPhone 13, 390x844):

• 31 /api/plans/:id/activity calls all return 200
• 24 chips show ⚪ paused, 7 show 🟢 done
• a437 / a447 → 🟢 done
• a433 → ⚪ paused (1d ago, 7.6h since last real heartbeat)
• Detail page activity card renders state + explanation

Files touched:

• `src/mavis-hq.js` — 1 new endpoint, 1 keep_alive filter block (~80 lines)
• `public/mavis-hq/index.html` — moved state-chip forEach to after innerHTML (~10 line move)

Commits:

• `da8b94e fix(plans): ship /api/plans/:id/activity endpoint (404 → 200)`

Bulletins:

• b442 — state chips work