Fix session start subscription ordering tests

[stephentoub]

Managed settings resolution can legitimately emit a live, ephemeral event before session.start. The C# and Python E2E tests assumed the start event would always be the first subscription callback, causing the CAPI test failure reported in github/copilot-agent-runtime#21956.

This updates the live-subscription tests to wait specifically for SessionStartEvent/SessionStartData while preserving their preallocated and server-assigned session ID assertions. Persisted event-log ordering assertions remain unchanged.

Generated by Copilot

[Copilot]

Copilot review overview

🟢 Approval recommended

The changes consistently remove the invalid first-event assumption while retaining the intended session ID checks.

Review effort: Balanced
Findings: None

What changed in this PR

Updates Python and .NET E2E tests to tolerate ephemeral events preceding session-start callbacks.

Changes:

• Wait specifically for session-start events.
• Preserve session ID and event-data assertions.
• Rename tests to reflect updated semantics.

File	Description
python/​e2e/​test_scenario_session_setup_e2e.py	Filters for preallocated session-start events.
python/​e2e/​test_scenario_cloud_e2e.py	Filters for cloud session-start events.
dotnet/​test/​E2E/​ScenarioTestingSessionSetupE2ETests.cs	Waits specifically for SessionStartEvent.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[github-actions]

SDK Consistency Review

This PR (#2737) fixes a race-condition assumption in the .NET and Python E2E tests: managed settings resolution can emit a live ephemeral event before session.start, so the tests previously assumed the very first subscription callback would always be the start event. The fix makes both tests wait specifically for SessionStartEvent/SessionStartData instead of the first event received, while preserving the preallocated- and server-assigned-session-ID assertions. Persisted event-log ordering assertions are untouched.

Cross-SDK check: I compared the equivalent scenario/E2E tests in Go, Rust, Node.js, and Java:

• go/internal/e2e/scenario_testing_cloud_e2e_test.go and go/internal/e2e/session_config_e2e_test.go already assert specifically on SessionStartData/SessionEventTypeSessionStart rather than assuming the first received event is the start event.
• rust/tests/e2e/client_options.rs (cloud_create_routes_first_event_for_server_assigned_session_id) already filters for SessionEventType::SessionStart / SessionStartData rather than relying on "first callback = start event".
• No equivalent bug pattern (blind first-event assumption) was found in the Java or Node.js E2E suites.

So this is a targeted bugfix that only needed correction in the .NET and Python test harnesses — the other SDKs' tests were already written correctly and don't need changes. No cross-SDK inconsistency is introduced by this PR. ✅

Generated by SDK Consistency Review Agent for #2737 · copilot · sonnet50 · 60 AIC · ⌖ 11.5 AIC · ⊞ 7.8K · ◷