test(mcp): preserve startup failure diagnostics

[Aaronontheweb]

Summary

• assert the MCP connection contract before reading the notification lease
• forward full manager exceptions to test output and cover failed stdio startup diagnostics

Evidence and scope

Windows run 31560045253, job 94000330013, failed after StartAsync because the test read a null NotificationLease. The test used a null logger and did not call the existing AssertConnected helper, so the run discarded the startup exception that would identify the production cause.

This change does not alter MCP runtime behavior, timeout values, retries, or lease assertions. It makes any connection failure fail first with the server state and status message. It also retains the manager's full exception in test output. The negative test proves that a missing stdio process reports Unreachable and Failed to reach MCP server, and that its command and transport exception reach the output, before the lease assertion.

This PR does not prove or fix an underlying production MCP startup defect. If the intermittent Windows startup failure remains, the next run will retain the exception needed for a causal follow-up.

Verification

• focused MCP notification and manager tests: 22 passed
• complete Netclaw.Daemon.Tests Release suite: 1,019 passed, repeated 3 times
• dotnet build Netclaw.slnx -c Release --no-restore: passed with 0 warnings and 0 errors
• changed-file dotnet format --verify-no-changes: passed
• pwsh ./scripts/Add-FileHeaders.ps1 -Verify: passed
• changed-file Slopwatch: 0 findings

The repo-wide Slopwatch scan still reports one pre-existing warning in PowerShellHostProbeTests.cs; this change does not touch that file.