fix: catch ClosedResourceError in error handler to prevent crash on client disconnect

When a client disconnects during request processing, _handle_message
receives the exception and tries to send a log message back to the
client via send_log_message(). Since the write stream is already closed,
this raises ClosedResourceError (or BrokenResourceError), which is
unhandled and crashes the stateless session with an ExceptionGroup.

Wrap the send_log_message() call in a try/except that catches both
anyio.ClosedResourceError and anyio.BrokenResourceError, since failing
to notify a disconnected client is expected and harmless.

This is a different code path from PR #1384, which fixed the message
router loop. This fix covers the _handle_post_request → _handle_message
→ send_log_message error recovery path.

Fixes #2064

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: giulio-leone

src/mcp/server/lowlevel/server.py

@@ -417,11 +417,16 @@ async def _handle_message(
                         )
                 case Exception():
                     logger.error(f"Received exception from stream: {message}")
-                    await session.send_log_message(
-                        level="error",
-                        data="Internal Server Error",
-                        logger="mcp.server.exception_handler",
-                    )
+                    try:
+                        await session.send_log_message(
+                            level="error",
+                            data="Internal Server Error",
+                            logger="mcp.server.exception_handler",
+                        )
+                    except (anyio.ClosedResourceError, anyio.BrokenResourceError):
+                        logger.warning(
+                            "Could not send error log to client: connection already closed"
+                        )
                     if raise_exceptions:
                         raise message
                 case _: