Fix inference step logs and cumulative reward clamps to ensure strictly (0, 1) bounds without 0.00 reporting

fix: comprehensive Phase 2 score range audit - eliminate all 0.0/1.0 leaks

fix(grader): keep scores within 0.05 and 0.95

fix(grader): explicitly clamp scores to [0.01, 0.99] as required by validator bounds

Clamp validator-visible scores to open interval

chore: verified code - zero modifications, all tests passing

last commit