fix: clamp all scores in grader AND inference to (0,1) exclusive

fix: clamp scores to (0,1) exclusive range for validator

fix: use [START]/[STEP]/[END] format with flush=True

fix: aggressive speed optimization - sample 5 attacks/queries, concurrent eval, timeouts

fix: add server/app.py and uv.lock at repo root

fix: add pyproject.toml at repo root

fix: add openenv.yaml at repo root for validator

fix: add inference.py at repo root

fix: make reset body optional for validator compatibility

fix: add START/STEP/END structured logging, fix env var defaults per submission checklist

Revert "fix: sample attacks and increase delays for rate limit compliance"

fix: sample attacks and increase delays for rate limit compliance

fix: sequential grading with retry logic for rate limits

fix: loosen utility checks and add rate limit handling

Add PromptGuard environment

initial commit