Update openenv test cases

Update groq model

Replace algorithmic scoring with Groq contextual LLM moderation

Adjust moderation thresholds for stricter tiering

Refactor moderation logic to use local scoring system

Fix hf_moderate error handling & loading fallback

Update deprecated Hugging Face inference pipeline endpoint to HF Router URL mapping

Extract upstream Hugging Face API errors natively dropping into UI fallback wrapper

Implement robust JSON safety fallbacks blocking malformed Mistral outputs from 500 downstream crashes

Swap moderation endpoint from OpenAI to free HuggingFace API

Raise transparent errors instead of passing silent ones on LLM loop bypass

Lazy load OpenAI client

Fix OpenEnv mapping architecture matching explicit tasks

Implement two-stage AI moderation pipeline in server/app.py

Make API Key private