Read API vars fresh in main(), add stderr debug logging

Fix API key priority: use API_KEY before HF_TOKEN

Rewrite inference.py to use OpenEnv client and BENCHMARK_URL

Add [START]/[STEP]/[END] structured output to inference.py

Improve inference strategy: filter flagged columns, cap plan size

Fix all code review findings (5 issues)

Transform to competition-grade submission

Improve inference agent for better scores

Fix stateful HTTP endpoints and add HF Spaces deployment config

Add inference script, Dockerfile, config, and README