Major env overhaul: opponent negotiates naturally, gentler time penalty, relative aggression, simplified agent

Smart offer clamping + accept guard + softened time penalty + adaptive concession cap

Refined prompt: multi-round negotiation with valuation guardrail, temp 0.3

Optimize baseline prompt: multi-round negotiation strategy, all 3 tasks now pass

Fix [END] format: restore score= field, add flush=True to all log prints

Fix inference.py stdout format: strict [START]/[STEP]/[END] compliance, all debug to stderr, try/finally for [END]

feat: complete OpenEnv compliance, tasks, and logic fixes

feat: implement negotiation environment wrapper and LLM-based inference script