test(mcp): surface stdio connect diagnostics in prompt smoke test

[Aaronontheweb]

Fixes the opaque Windows failure of SmokeMcpPromptSkillTests.ManagerDiscoversAndLoadsPromptOverStdio(serverKind: \"python\").

Problem

On Windows CI (run 31794198529), the python theory case fails after 1m10s with Assert.IsType() Failure: Value is null at line 37. The manager swallows the MCP SDK TimeoutException (60s InitializationTimeout default) into an Unreachable status, and this test never calls AssertConnected nor passes an ITestOutputHelper — so CI shows only an opaque null and the real startup error goes to NullLogger.

Prior context

• PR test(mcp): assert smoke harness connectivity deterministically #1832 added AssertConnected + test logger but applied it only to SmokeMcpServerHttpHeaderTests.
• PR test(mcp): preserve startup failure diagnostics #1904 fixed the identical failure shape on McpSdkCatalogNotificationIntegrationTests with an explicit note: "This PR does not prove or fix an underlying production MCP startup defect... the next run will retain the exception needed for a causal follow-up." That retention was never applied to this test.

Fix

Diagnostics-first, per the #1904 pattern — no timeout jiggling:

• Inject ITestOutputHelper into the test.
• Call harness.AssertConnected(\"smoke\") before the skill lookup, so the next Windows failure names the manager's real exception instead of an opaque null.

This is the prerequisite for a causal fix (expected: python stdio child exceeds the 60s init budget on Windows). If the retained error confirms a timeout, the follow-up is to raise the stdio InitializationTimeout in McpClientManager.BuildClientOptions.

[Aaronontheweb]

This is just a test observability fix, no actual behavioral or test changes yet.