fix: prevent After(TestSession) hook from firing prematurely on test timeout

[thomhurst]

Summary

• Adds a first-call-wins guard to AfterHookPairTracker.RegisterAfterTestSessionHook so only the initial registration (using the session-level cancellation token) takes effect — subsequent calls from per-test timeout tokens are ignored
• Routes TestScheduler's post-execution session hook call through AfterHookPairTracker.GetOrCreateAfterTestSessionTask instead of calling HookExecutor directly, ensuring hooks execute exactly once even if a cancellation callback already triggered them
• Wires up the AfterHookPairTracker dependency in TUnitServiceProvider

Test plan

• Build succeeds (dotnet build TUnit.sln)
• Verify with reproduction case from [Bug]: [After(TestSession)] hook incorrectly triggered on test timeout #4712: timeout test + DependsOn + After(TestSession) hook fires exactly once
• Existing engine tests pass with no regressions

Closes #4712

🤖 Generated with Claude Code

[claude]

Performance optimization opportunity in TestScheduler.cs line 143: The lambda creates an unnecessary ValueTask to Task to ValueTask conversion. Since GetOrCreateAfterTestSessionTask internally calls .AsTask() on the returned ValueTask, you can simplify the lambda to just: () => _hookExecutor.ExecuteAfterTestSessionHooksAsync(cancellationToken). This eliminates one allocation cycle and aligns with CLAUDE.md Rule 4 about minimizing allocations and using ValueTask appropriately. See https://github.com/thomhurst/TUnit/blob/0075a0134d8c7c5ed58d9d5820df1e20f7441293/CLAUDE.md for the rule and

TUnit/TUnit.Engine/Services/AfterHookPairTracker.cs

Lines 108 to 110 in 0075a01

	{
	_afterTestSessionTask = taskFactory().AsTask();
	}

for the .AsTask() call.