fix: correct RL reward signal, disable reactive red pivot, propagate negligence penalty to dimensions

Fix 4 critical RL training bugs

Add hotseat multiplayer Red Team controls and 4 architectural fixes

Gracefully handle invalid LLM actions with -0.2 penalty instead of 500 crash

Fix attack_chain KeyError and other validation errors

Initialize _step_reward_total in __init__ to fix AttributeError

Add root health check route to fix HF 404

Add FSP multi-agent architecture: Red Team LLM action space + alternating turns

Fix business_impact grader: penalise unjustified isolations per-host

Add alternating self-play training scaffolding.

Replace SOAR playbooks with agentic micro-tools.

Add .gitignore, improve play_environment pending_followup tracking

Fix phantom data bug: correct HF baseUrl, health endpoint for liveness probe, zero initial scores

Initial commit of CyberSOC upgraded RLVR environment