Stabilize MCP lifecycle tests under suite load

[Widthdom]

Summary

• Make MCP batch timeout tests establish initialization outside the shortened timed dispatch.
• Synchronize timeout-action release on observed cancellation and use typed response-shape assertions.
• Replace the cancelled EOF drain wall-clock threshold with a direct cancellation contract.
• Use a database-independent adjacent success request in the duplicate-initialize timeout fixture.

Root cause

• The shortened RequestTimeout also covered setup initialization and DB-probing ping requests, so parallel-suite load could change the expected lifecycle or response shape.
• The timeout fixtures released blocked actions without first proving timeout cancellation had been observed.
• The EOF coverage used a host-load-sensitive elapsed-time threshold instead of asserting the actual cancellation behavior.

Validation

• dotnet build CodeIndex.sln -c Release -p:UseSharedCompilation=false --no-restore — succeeded with 0 warnings and 0 errors.
• Focused MCP tests — 3/3 passed on both net8.0 and net9.0 in Release.
• Adversarial stress — the three focused tests passed 30/30 iterations on net9.0.
• dotnet run --project tools/CodeIndex.Changelog -- check — validated 95 fragments.
• Local CodeIndex status --check — indexed HEAD matched the clean workspace with all readiness checks passing.
• Codex adversarial review — No blocking/actionable issues found.

Documentation and changelog

• Updated the English and Japanese testing contracts in TESTING_GUIDE.md.
• Added bilingual internal fragments for Stabilize MCP batch timeout concurrency lease test under full-suite load #4935, Stabilize the batch timeout concurrency test under parallel suite execution #4947, Stabilize cancelled EOF drain timing on .NET 9 #4978, and Stabilize net9 MCP lifecycle tests under full-suite load #5001.

Follow-up candidates

• None.

Fixes #4935
Fixes #4947
Fixes #4978
Fixes #5001