fix(mcp): ensure server.shutdown() on probe iteration failure

Wrap the _tools iteration in _probe_single_server() in try/finally
so that server.shutdown() is called even if iterating tool metadata
raises. Without this, the MCP server connection leaks until the
event loop is torn down by _stop_mcp_loop().

hermes_cli/mcp_config.py

@@ -225,13 +225,15 @@ def _probe_single_server(
         server = await asyncio.wait_for(
             _connect_server(name, config), timeout=connect_timeout
         )
-        for t in server._tools:
-            desc = getattr(t, "description", "") or ""
-            # Truncate long descriptions for display
-            if len(desc) > 80:
-                desc = desc[:77] + "..."
-            tools_found.append((t.name, desc))
-        await server.shutdown()
+        try:
+            for t in server._tools:
+                desc = getattr(t, "description", "") or ""
+                # Truncate long descriptions for display
+                if len(desc) > 80:
+                    desc = desc[:77] + "..."
+                tools_found.append((t.name, desc))
+        finally:
+            await server.shutdown()
 
     try:
         _run_on_mcp_loop(_probe(), timeout=connect_timeout + 10)