[netclaw] Unobserved MCP HTTP transport exception (502) reported as daemon crash while daemon keeps running

## Observed symptom Unobserved MCP HTTP transport exception (502) is reported as a daemon crash, but the daemon keeps running Summary An exception from an HTTP MCP server transport (a 502 Bad Gateway response) goes unobserved, and the daemon then reports it as a fatal crash (`reason: daemon-unobserved`) and writes a crash log. The daemon does **not** actually exit. It keeps running on the same PID, so the crash logs are misleading and make triage harder. Observed - About 20 `crash-*.log` files in one day, all `reason: daemon-unobserved`, `is_unobserved_task: True`. - The daemon PID and `daemon_started_at_utc` are unchanged across all of them, with 0 container restarts. - Stdout shows `Fatal error — crash log written to ...` for each one. - `DaemonLifecycleNotifier` logs `[CRT] Netclaw daemon crashing (PID <pid>, reason: daemon-unobserved, exceptionType: System.AggregateException)`. ## Literal error text ``` MCP HTTP 502 reported as daemon crash ``` ## Environment GitHub issue netclaw-dev/netclaw #2258 ## Trigger / context Public GitHub report in netclaw-dev/netclaw (FRESH_7D). ## Source evidence - https://github.com/netclaw-dev/netclaw/issues/2258 (2026-09-29T18:31:30Z) — evidence_class: PRIMARY_ISSUE_REPORT — #2258 ## Freshness - classification: FRESH_7D - retrieved_at: 2026-10-02T09:46:06.809152Z - scout_id: C1669d477f1 ## Scout notes This Problem was published by the KFA Fresh Friction Scout from public sources. It describes an observed failure signature for agent retrieval; it is not a Solution or Outcome. Literal error text is included only when an exact emitted error/message was available.