Fix live execution reconnects after dashboard navigation

[mbakgun]

What changed

• Preserve the owning workflow/execution state when an in-progress canvas run navigates through the dashboard and returns via History → Open live.
• Keep completed node results and runtime node status available when the same workflow canvas reloads.
• Send SSE heartbeat comments from the live execution observer during long-running idle nodes.
• Retry the live observer after transient stream disconnects, replaying the latest snapshot and falling back to persisted history on completion.
• Extend Playwright coverage for the exact canvas → dashboard → History → live canvas flow and add a backend heartbeat regression test.

Root cause

Leaving the editor cleared currentExecutionId while the original execution stream and abort controller remained active. Returning through Open live started a second observer and aborted the original stream, which could cancel or orphan the request-owned execution lifecycle. The observer stream also emitted no idle heartbeat and treated a dropped connection as terminal.

Impact

Long-running nodes continue to advance and finalize on the canvas after dashboard navigation. Live observers remain connected through idle periods and recover from transient SSE disconnects.

Validation

• SECRET_KEY=test-secret-key-for-tests-only-32-bytes ./check.sh — 2,296 backend tests plus frontend lint/typecheck passed.
• ./run_e2e.sh e2e/execution-deep-link.spec.ts --grep "keeps a canvas run live" — passed.
• ./run_e2e.sh e2e/execution-deep-link.spec.ts --grep "opens one running execution live" — passed.