fix(autoresearch): stream GAN output live

[FluffyAIcode]

Summary

• replace hidden capture_output GAN execution with line-buffered live streaming
• preserve Generator and Critic token output exactly
• preserve Prefill progress, percentage, and ETA
• preserve Goal Alignment and KV/decode/latency/E2E metrics
• retain full output internally for run ID parsing and evaluator input
• enforce the existing experiment timeout with a hard process-kill watchdog

Root cause

The supervisor captured the entire REPL subprocess in memory, so the original Terminal showed only AutoResearch phase lines until the experiment ended. Inference was running but all established GAN observability was hidden.

Tests

• supervisor source contract rejects capture_output and requires Popen live tee
• existing supervisor logic tests — 5 passed
• full platform-neutral suite — 890 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.