Stop running trace when starting a new trace

Is your feature request related to a problem? Please describe.

In an agentic LLM workflow, every step counts towards increasing the load on the AI and its context window. Performance traces are a particularly sore point for LLMs, as only one can be active at any given time. Currently, the MCP server throws an error if attempting to start a new trace when one already exists, even if that trace was from an entirely different conversation completely unrelated to the present one.

Describe the solution you'd like

Rather than throwing an error and interrupting the thought processes of an LLM when running performance_start_trace, close any existing performance trace and include a notice that this action happened in the response given by the start tool.

Describe alternatives you've considered

N/A

Additional context

No response

Start at the MCP server's performance_start_trace tool and trace how it handles an already-active performance trace. The completed behavior should close the existing trace, start the new one, and report that closure in the tool response instead of returning an error; verify the existing-trace scenario.