Preserve inference runtime across AutoResearch iterations

[FluffyAIcode]

Summary

• remove all worker deployment, launchctl restart, and automatic cache clearing from the decomposition supervisor
• replace lifecycle mutation with read-only Primary/allens/dashboard health checks
• keep worker chunk configuration immutable and move cold benchmarking outside the proof loop

Test plan

• targeted AutoResearch tests: 39 passed
• full pytest: 1698 passed, 60 failed, 2 skipped; failures match existing Qwen3/Transformers runtime incompatibility
• source guard verifies supervisor contains no deploy_candidate, clear_primary_cache, launchctl, or bootout

The uncommitted candidate.py file is active runtime state and is excluded.

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.