MCP connection leaks closed after webhooks

Webhooks that use MCP client tools were leaking connections on every request. The webhook handler received a list of cleanup callbacks but never executed them after returning—a classic resource management oversight. Each time a webhook ran, a new outbound SSE or HTTP connection opened to the MCP server and never closed.

Two specific scenarios caused this: when the webhook handler completed (regardless of success or error), close functions were ignored entirely; and when an MCP server returned no tools, the client connection was left hanging. This PR ensures cleanup code always runs by draining the close functions in a finally block, closing the client explicitly when the server has nothing to offer, and wrapping tool construction in error handling that guarantees the client shuts down cleanly.

The fix lives in the CLI webhook service and the MCP client tool node in the n8n LangChain integration. Connections to MCP servers are now managed correctly whether webhooks succeed, fail, or return empty tool lists.