Re-apply training prep work on top of validated Layers 1-5 base

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