Preserve task continuity across transient reconnects

[wolfiesch]

Summary

• Split reconnect and reconciliation changes out of Add safe local extension deployment and health checks #54.
• Preserve active ownership while ignoring stale tabs belonging to closed or missing tasks.
• Retry transient endpoint failures within the connection deadline without replaying dispatched handshakes.
• Prove a real transient failure occurred before the successful retry in the TypeScript regression.

Verification

Workspace typechecks, SDK build, TypeScript SDK contracts, and adapter reconnect contracts passed. Python and Rust contracts passed on the combined implementation; exact-head CI covers this focused branch.

[omp-maintainer]

review:p0 — scoped continuity fix: active ownership survives stale inventory, and endpoint retries stop before handshake replay.
One should-fix test-harness issue: the Python no-replay worker can survive its test after an unchecked timed join.
The two targeted Python regressions pass; Rust, TypeScript, and OMP contracts could not run here because cargo, tsc, and typebox are unavailable.
Thanks for the focused split.