fix(dashboard): send periodic WS heartbeat pings to stop live-dashboard reconnect churn

[diegosouzapw]

Closes #10319

Root cause

Confirmed in two files (matches the analyzed plan _tasks/pipeline/bugs/2-implementing/10319-live-ws-client-never-sends-heartbeat-ping.plan.md):

• Client (src/hooks/useLiveDashboard.ts): ws.onopen sent only { type: "subscribe", channels }. No setInterval/keepalive timer existed anywhere in the hook, so the client never emitted the protocol's declared { type: "ping" } heartbeat frame.
• Server (src/server/ws/liveServer.ts): client.lastActivity was refreshed only from inbound client messages. startHeartbeat terminates any client whose inbound gap exceeds HEARTBEAT_TIMEOUT_MS (35s, hardcoded). The server's own outbound { type: "pong" } did not reset lastActivity (inbound-only), so it could not prevent the terminate.

Net effect: a healthy, connected-but-idle dashboard client (the normal case — the dashboard just watches, sends nothing) was force-terminated (ws.terminate(), abnormal close) roughly every 35-45s, and the browser auto-reconnected in a loop — exactly the reporter's symptom (wss://…/live-ws failed: The network connection was lost every ~35-45s).

Fix (both directions, per the plan's recommendation)

1. Client — useLiveDashboard.ts now starts a 15s setInterval on open that sends { type: "ping" }, cleared on every exit path (close, unmount, disable, reconnect) to avoid leaked timers. This hook is the shared base for useLiveRequests, useLiveComboStatus, useLiveConnectionStatus.
2. Server (defense in depth) — startHeartbeat now also bumps client.lastActivity when sending the outbound pong, so even a third-party/bespoke client that never pings is not dropped for being idle.

Regression tests

• tests/unit/useLiveDashboard-heartbeat.test.tsx (vitest, fake timers): asserts the hook emits periodic { type: "ping" } frames after open, and that the interval is cleared on close and on unmount (no leaked timers).
• tests/integration/live-ws-heartbeat-keepalive.test.ts (node:test, real WS server via scripts/start-ws-server.mjs): a subscribed-but-silent client must stay connected past the 35s HEARTBEAT_TIMEOUT_MS (~50s window) — the permanent version of the plan file's TDD RED repro, which failed before this fix (AssertionError: true !== false, connection was terminated at ~44s).

TDD RED→GREEN evidence is documented in the analyzed plan file (verdict: fix-confirmed, confidence: CONFIRMED). Pure protocol defect, not environment-specific — no VPS live validation required per the plan.

Gates run locally

• npm run typecheck:core — GREEN
• npx eslint --suppressions-location config/quality/eslint-suppressions.json <changed files> — GREEN, 0 errors
• node scripts/check/check-file-size.mjs — no new violations on touched files
• node scripts/check/check-complexity.mjs — GREEN (2456 vs baseline 2774)
• node scripts/check/check-cognitive-complexity.mjs — GREEN (1104 vs baseline 1223)
• node scripts/check/check-test-discovery.mjs — GREEN, new test files discovered by the runner

Note: npm run test:unit and the targeted npx vitest run tests/unit/useLiveDashboard-heartbeat.test.tsx could not be completed locally in this session — this devbox is running a 13-way parallel campaign fan-out and sustained a load average of ~150-240 on 16 cores for 30+ minutes; the local test:unit process measured 0.0% CPU over 26 minutes of wall time (confirmed via ps), and vitest's worker-pool spawn repeatedly hit its internal startup timeout under the same contention ([vitest-pool-runner]: Timeout waiting for worker to respond — an infra symptom, not a test assertion failure). Both new tests were manually verified line-by-line for correctness against the existing tests/integration/live-ws-startup.test.ts harness pattern and the hook's actual timer/cleanup logic. CI's test-unit and test-vitest jobs will run the authoritative suite on this PR.

⚠️ base-red inherited: #9985 — ESLint errors (2) from #10250 (i18n PT-PT translation), unrelated to this change.