Python: Propagate MCP initialization failures

[lxingy3]

Motivation and Context

Closes #13414.

When MCP streamable HTTP session initialization fails, for example after an authentication error from the server, the background connection task can raise before ready_event is set. connect() keeps waiting on that event, so the caller sees a hang instead of the initialization error.

Description

This change makes the connection setup fail closed:

• _inner_connect() sets ready_event before propagating any setup failure.
• connect() checks whether the background task already completed after the event is set, then awaits it so the original KernelPluginInvalidConfigurationError reaches the caller.
• A unit test covers a streamable HTTP connection where transport setup succeeds but ClientSession.initialize() fails.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the checked files pass ruff
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone

Validation:

• python -m pytest python\tests\unit\connectors\mcp\test_mcp.py::test_streamable_http_initialization_failure_unblocks_connect -q
• python -m pytest python\tests\unit\connectors\mcp\test_mcp.py -q
• python -m ruff check python\semantic_kernel\connectors\mcp.py python\tests\unit\connectors\mcp\test_mcp.py
• git diff --check

[lxingy3]

@microsoft-github-policy-service agree