Run Tools in Parallel, With Timeouts

A model asks for three tools in one turn. Run them together, and survive the one that never returns.

Implement run_parallel(tools, calls, timeout=0.5).

• tools maps a name to a function; calls is a list of {"name": ..., "arguments": {...}}.
• Run every call concurrently; concurrent.futures.ThreadPoolExecutor is enough.
• Return one result per call, in the order the calls were given, each {"name": ..., "content": str, "is_error": bool}.
• A tool that raises: is_error: True, content is the error message. One failure must not lose the others.
• A tool that exceeds timeout: is_error: True, content mentions the timeout. The others still return.
• An unknown tool name: is_error: True, content says it is unknown, never submitted.

The catch: concurrent work finishes in any order, and a result in the wrong slot answers the wrong question. Submit in order, collect in order.

The tests time a run: two tools that each sleep 0.3s must finish together in well under 0.6s.