fix: match reference project structure to resolve platform validation

Fix: inference.py always exits 0, fixes chain-gate loop, adds full traceback logging

Fix: use 'or' fallback for MODEL_NAME/API_BASE_URL to handle empty-string env vars from evaluator

Fix false-positive env var validation: check resolved Python vars, not os.environ directly

Set default API_BASE_URL and MODEL_NAME to match sample_interface.py

fix: inference.py stdout format compliance

fix: align step_reward with grade_episode, pin deps, update docs, clean inference

fix: resolve all 3 pre-submission critical risks

feat: enforce investigation discipline + fix easy-task grading + add investigation_hints

feat: reproducible baseline — fixed seed, correct_action in table, accuracy summary

fix: per-task reward display uses weighted_reward key from grader

feat: run LLM inference baseline + fix SSL and loop guard in inference.py

feat: Iteration_3 — openenv validate PASS, wire compat, PayOpsReward, OPENAI_API_KEY- Fix completed, Docker verification alone pending- Padma

chore: pre-iteration-1 snapshot — 75/75 e2e passing, baseline bug fixed