implemented agents' self-learning, self-correcting without explicit training

modified inference to include [START] at the beginning of each task

modified inference to include proper prints

created tasks.py, dated env, inference

updated environment to get tasks from graders

fix: handle step before reset (state_data None)

modified Dockerfile, inference.py, app/env.py; included openenv.yaml, server7860.py

fresh clean commit