feat: add grader classes for all 3 DRL tasks

feat: implement state interface, standardize schema IDs, and add task discovery endpoint

fix: set default value for ResetRequest in reset_env endpoint to prevent validation errors

feat: implement reinforced human memory for inference and enhance UI with audit capabilities and task labeling

feat: enhance safety moderation with keyword-based fallback, robust LLM response parsing, and unified UI insight formatting.

refactor: update submission validator with strict log format checks, HF_TOKEN safety verification, and add FastAPI server for interactive task testing