[everpcpc]

Summary

• Add boot-scoped, monotonically increasing IDs to domain events and retain the most recent 1024 events in memory.
• Resume native EventSource connections with Last-Event-ID only when the complete event sequence is still available.
• Start with live events without replay when continuity cannot be proven; no custom recovery event is added to the SSE protocol.
• Keep the web UI unchanged: it applies events as they arrive and does not automatically refresh all state after a reconnect.

Recovery semantics

• Subscribing to live events and choosing whether to replay are atomic, so events cannot fall between replay and live delivery.
• Replay preserves the existing administrator and per-user event scopes.
• Heartbeats and TaskQueueStatus snapshots do not advance the recovery cursor.
• A lagging stream closes so the browser can reconnect and either replay or continue live.
• Replay is best-effort: events outside the retained history and one-off notifications during a gap can be missed. Users can reload the page if they notice stale state.
• Recovery is limited to the current browser connection lifecycle; no event IDs are persisted across page loads.

Validation

• cargo nextest run --workspace — 1226 passed, 1 skipped
• cargo fmt --all -- --check
• cargo clippy --workspace --all-targets -- -D warnings
• corepack pnpm build and corepack pnpm lint in webui/
• Regression tests were shown to fail before the corresponding replay and recovery changes.
• Browser verification: after a server restart and out-of-band database rename, the open page kept showing Before Reconnect; after a manual reload, it showed After Reconnect.

Documentation

• docs: document SSE recovery website#63 documents complete-only replay, the live fallback for incomplete history, and the best-effort limits.