Fix connection closure handling in Protocol and InMemoryTransport

[alasano]

Fix connection closure handling in Protocol and InMemoryTransport

Motivation and Context

This PR fixes two critical bugs that cause promises to hang indefinitely when connections are closed:

1. InMemoryTransport deadlock: When closing one transport, the peer's onclose callback was never fired, causing server-side promises to hang forever
2. Protocol request handler leak: Active request handlers were not aborted on connection close, causing them to continue running indefinitely

These bugs manifested as timeout errors in applications where connections were closed while requests were in-flight (e.g., user disconnection during a pending UI prompt).

How Has This Been Tested?

• Added comprehensive unit tests for InMemoryTransport close behavior (idempotency, concurrent closes, post-close operations)
• Tested with real-world scenario: closing connection while server has pending elicitation/create request
• All existing tests pass
• Verified fix resolves timeout issues in a multi-user MCP client implementation where disconnecting users with pending server-initiated requests (e.g., elicitation/create) would cause tests to hang indefinitely

Breaking Changes

None. These are bug fixes that make the SDK behave as expected.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update

Checklist

• I have read the MCP Documentation
• My code follows the repository's style guidelines
• New and existing tests pass locally
• I have added appropriate error handling
• I have added or updated documentation as needed

Additional context

The fix ensures proper cleanup in both directions:

• InMemoryTransport now guarantees both transports fire their onclose callbacks
• Protocol now aborts all active request handlers via their AbortControllers
• Together, these changes prevent resource leaks and hanging promises during connection closure

[alasano]

Hey @ihrpr , this PR fixes critical bugs causing indefinite connection hangs in multi-user scenarios.

The fixes are also non-breaking with proper test coverage.

Could you please review when you have time? Or assign someone else if possible, we've been running our own version with the fix but need it in the official package ideally.

[ihrpr]

Thank you so much for the fixes!

Please can tests be added for the Protocol abort functionality as well?

[alasano]

Thank you so much for the fixes!

Please can tests be added for the Protocol abort functionality as well?

Thank you for getting back to me as well 🙏

Absolutely, I'll add some a bit later today

[markdascher]

Would #611 be fixed by this PR?

[felixweinberger]

Hi @alasano, re-submitted this after verifying it's still a problem at #1735 + added tests, Co-authored-by: you, so closing this as superseded.