fix: handle stale CLI process error during worker session resume

[PureWeen]

Problem

When the CLI server process dies, the SDK's \StartCliServerAsync\ calls \Process.HasExited\ on a stale handle, throwing:
\
System.InvalidOperationException: No process is associated with this object.
\
This error was not recognized by \RestorePreviousSessionsAsync's fallback logic, so worker sessions (e.g., \iesta-worker-1) were silently dropped during resume — causing orchestration to fail.

Root Cause

The catch block in \RestorePreviousSessionsAsync\ only handled three error patterns: \Session not found, \corrupt, and \session file. The process-related \InvalidOperationException\ matched none of them and also wasn't classified as a connection error by \IsConnectionError().

Fix

• **\CopilotService.Utilities.cs**: Add \IsProcessError()\ helper that detects stale process handle exceptions (\InvalidOperationException\ with \No process is associated). Include it in \IsConnectionError()\ so retry/reconnect logic works across all call sites.
• **\CopilotService.Persistence.cs**: Add \IsProcessError(ex)\ to the fallback condition in \RestorePreviousSessionsAsync\ so affected sessions get recreated via \CreateSessionAsync\ instead of being silently dropped.

Tests

Added 9 regression tests in \ConnectionRecoveryTests.cs:

• \IsProcessError\ detection (direct, AggregateException-wrapped, inner exception)
• \IsProcessError\ negative cases (unrelated InvalidOperationException, non-IOE)
• \IsConnectionError\ process handle detection (direct, wrapped, inner)
• Structural guard verifying the restore fallback includes \IsProcessError\

All 2570 tests pass. Windows build succeeds.

[PureWeen]

PR #374 Review — fix: handle stale CLI process error during worker session resume
CI: ⚠️ No checks configured
Consensus Findings (2+ of 5 models)

Sev Location Finding Consensus

1 🟡 ConnectionRecoveryTests.cs:~444 Structural test uses Substring(conditionIndex, 500) — fragile fixed window that can ArgumentOutOfRangeException or false-fail if code shifts. Tests source text, not behavior. 5/5
2 🟡 ConnectionRecoveryTests.cs No behavior-level test proving RestorePreviousSessionsAsync actually takes the CreateSessionAsync fallback when the process error fires. 2/5
3 🟡 CopilotService.Utilities.cs:338 "No process is associated" is an English BCL string — fragile on localized runtimes (low real-world risk on .NET Core, but worth noting). 2/5
4 🟢 CopilotService.Utilities.cs:323-327 Redundant double-traversal of AggregateException inner exceptions — IsProcessError walks them, then IsConnectionError walks again. Correct but wasteful. 3/5
5 🟢 ConnectionRecoveryTests.cs:~444 conditionIndex > 0 should be != -1 (IndexOf returns -1 on miss, 0 at position zero). 1/5 (noted)
Verified Clean
IsProcessError() correctly guards on type (InvalidOperationException) + message substring ✅
Fallback condition uses || correctly — process errors OR'd in, not replacing existing checks ✅
Exception chain traversal terminates correctly ✅
9 unit tests cover detection logic including negative cases ✅
Verdict: ⚠️ Request Changes
One ask: Replace or harden the source-scan structural test (RestorePreviousSessions_FallbackCoversProcessErrors):
At minimum: use Math.Min(500, source.Length - conditionIndex) and switch to IndexOf("IsProcessError", conditionIndex) instead of a fixed window
Ideally: add a behavior-level test that throws the process InvalidOperationException during restore and asserts the CreateSessionAsync fallback fires