fix(prefill): isolate progress HTTP event loop

[FluffyAIcode]

Summary

• run the inference-network Uvicorn server in its own thread and event loop
• keep gRPC model operations on their existing MLX-affine execution path
• allow heartbeat polling to read segment token progress while AppendTokens blocks
• shut down and join the HTTP thread during graceful runtime termination

Root cause

The worker correctly reported segmented tokens_computed, but Primary's synchronous AppendTokens path occupied the same asyncio loop that hosted Uvicorn. Dashboard requests timed out during Prefill, so Terminal heartbeats displayed stale Generator progress instead of live Critic segments.

Tests

• HTTP/gRPC event-loop isolation source contract
• targeted bridge/network suite — 20 passed
• full platform-neutral suite — 874 passed, 1 skipped
• Python compile and diff checks pass

Made with Cursor

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.