Add behavioral tests for orchestration recovery paths

[PureWeen]

Context

PR #375 added significant orchestration recovery logic (premature idle recovery, dead event stream fallback, lazy session resume). The test coverage is currently mostly structural (source-code pattern checks via Assert.Contains) rather than behavioral.

Missing behavioral test coverage

1. EnsureSessionConnectedAsync — no end-to-end test for resume + fresh-create fallback + GetClientForGroup routing
2. Premature idle recovery loop — no test with real ManualResetEventSlim signaling and OnSessionComplete events
3. bestResponse multi-round accumulation — no test verifying longest response wins across recovery rounds
4. DateTime filtering correctness — no test verifying dispatchTime filter works with local timestamps
5. OCE handling — no test verifying bestResponse is preserved when inner recovery CTS fires
6. Dead event stream recovery — no test for LoadHistoryFromDiskAsync fallback with temp events.jsonl

Suggested approach

Create integration-style tests using Demo mode stubs that simulate the event sequences (premature idle, dead streams, OCE). The MultiAgentRegressionTests pattern with StubCopilotSession could be extended.

Priority

Medium — structural tests catch regressions at the code-pattern level, but behavioral tests would catch logic bugs that structural tests miss.