Reset env between retry attempts within a single episode

Add inference-time retry loop with feedback for self-correction

Add in-context demonstration learning support

Fix make_plots.py to accept --wandb-run-id separately from --training-log-json

Add HF Job training pipeline: persistence-aware run script, judge-facing demo notebook, baseline JSON output

Initial commit: MutantHunter — RL env for mutation-score-rewarded test generation