fix: handle ClientDisconnect gracefully instead of returning HTTP 500

[skyvanguard]

Problem

StreamableHTTPServerTransport._handle_post_request (and related SSE handlers) incorrectly handles starlette.requests.ClientDisconnect exceptions. When a client disconnects during a request, the broad except Exception handler catches it and:

• Returns HTTP 500 (Internal Server Error)
• Logs as ERROR with full traceback via logger.exception()
• Triggers production 5XX alerts and monitoring noise

This happens during normal operations: network timeouts, user cancels request, load balancer timeouts, mobile client network interruptions. These are client-side events, not server failures.

Solution

Catch ClientDisconnect explicitly before the generic except Exception handlers:

1. handle_request() — wraps all HTTP method dispatching (POST, GET, DELETE) with a single except ClientDisconnect that logs at DEBUG level. This is the primary fix that prevents the 500 response for the most common case (client disconnects during body reading or response sending).

2. _handle_get_request() SSE handlers — catches ClientDisconnect in the standalone SSE writer and SSE response to cleanly close streams without ERROR logging.

3. Event replay handlers — catches ClientDisconnect in the replay sender, replay response, and outer replay handler.

All handlers log at DEBUG level since client disconnection is expected behavior, not an error condition.

Testing

Added regression tests in tests/issues/test_1648_client_disconnect_500.py:

• test_client_disconnect_does_not_produce_500 — verifies no ERROR-level logs are produced when a client disconnects during a request
• test_server_healthy_after_client_disconnect — verifies the server remains operational and can accept new requests after a client disconnects

Closes #1648