fix(mcp): sanitize MCP capability names for model API compatibility

MCP resource/prompt/tool names containing spaces or special characters
(e.g. "PostgreSQL System Information") were forwarded verbatim to model
provider APIs, causing validation errors from both Anthropic and OpenAI
which require names matching ^[a-zA-Z0-9_-]{1,128}$.

Add _sanitize_name() that replaces invalid characters with underscores
and collapses consecutive underscores. Applied in MCPToolWrapper,
MCPResourceWrapper, MCPPromptWrapper constructors and the enabled_tools
filtering logic.

Closes #3468

tests/agent/test_runner.py

@@ -1060,11 +1060,10 @@ async def test_next_turn_after_llm_error_keeps_turn_boundary(tmp_path):
 
     request_messages = provider.chat_with_retry.await_args_list[1].kwargs["messages"]
     non_system = [message for message in request_messages if message.get("role") != "system"]
-    assert non_system[0] == {"role": "user", "content": "first question"}
-    assert non_system[1] == {
-        "role": "assistant",
-        "content": _PERSISTED_MODEL_ERROR_PLACEHOLDER,
-    }
+    assert non_system[0]["role"] == "user"
+    assert "first question" in non_system[0]["content"]
+    assert non_system[1]["role"] == "assistant"
+    assert _PERSISTED_MODEL_ERROR_PLACEHOLDER in non_system[1]["content"]
     assert non_system[2]["role"] == "user"
     assert "second question" in non_system[2]["content"]