feat(dashboard): reconnect activity SSE with backoff and backfill missed events

[vreabernardo]

Closes #258

Stacked on #273 (main does not build at HEAD; this branches from the repaired tree). Happy to rebase once #273 merges.

Before this, a dropped SSE connection meant the feed silently fell back to the polling interval and anything published during the outage only showed up on the next poll. Now:

• The client retries the stream with exponential backoff and jitter (1s base, doubling, capped at 30s). Polling still takes over, but only after 5 consecutive failed attempts, and it still handles the EventSource-unavailable case immediately. NEXT_PUBLIC_ACTIVITY_REFRESH_MS=0 manual-only mode is untouched.
• Activity frames carry an id line, so native EventSource tracks lastEventId. The manual reconnect passes it along as ?lastEventId= and the stream route replays stored events published since that cursor (snapshot taken before subscribing, so nothing published in between is missed). Unknown cursor replays nothing, the REST backfill covers that.
• After a reconnect the feed hook backfills through the activity API with from=, sort=oldest before live delivery resumes. The from filter is inclusive and the hook dedupes by event id, so overlap with the server replay is harmless.

Tests: the three client-connector tests that asserted immediate fallback now cover the reconnect contract (backoff instead of a tight retry loop, cursor on the reconnect URL, exhaustion into polling after N attempts), plus new coverage for the server replay and getEventsAfterCursor. Ran the full dashboard suite (556 passed, 0 failed, tsx --test across test/**) and eslint on the touched files. tsc reports only the pre-existing baseUrl deprecation in tsconfig.json.

[vreabernardo]

Withdrawing this contribution. Thanks for your time.