v2.0: multi-step episodes, procedural bugs, semantic grading, sessions, 71 tests

Fix all things

fix: stateful endpoints + score clamping for validator pass

fix: replace 0.0 fallback with 0.05 in graders to satisfy strict range

fix: no exact 0.0 or 1.0 anywhere in rewards

final fix: validator-compatible graders

upgrade task.py: milestone grading, team in medium, 5 bugs per task

fix import paths for Docker

complete bug triage openenv environment