[HarpeLm]

Closes #1410

Retry transient persist_status() failures during the Server-owned Tunnel heartbeat with bounded 25/50/100 ms backoff. Persistent publication failures still terminate the profile with the explicit tunnel_health_unavailable diagnostic.

Validation:

• cargo test --locked -p webcodex --lib server_tunnels
• cargo fmt --all -- --check
• git diff --check

[swjnxyf]

Thank you for taking the time to investigate #1410 and contribute a focused fix. We appreciate the clear implementation and tests, especially the bounded-retry approach.

We've now merged #1411, which addresses the same heartbeat-persistence failure. We chose to keep an already-running Server-owned Tunnel alive through publication errors and retry on subsequent normal health ticks, rather than terminate the Tunnel if a short retry window is exhausted. This also adds deterministic contention/recovery coverage, including Windows CI.

Since the two changes overlap in the same code path, we'll close this PR as superseded by #1411 to avoid introducing competing recovery behaviors. Your short-backoff approach is still useful feedback and something we can revisit if needed.

Thanks again for contributing to WebCodex!