Use raw HF Inference API instead of InferenceClient provider system

Fix stop_sequences deprecation, expose full error traceback

Switch to text_generation — model not registered as chat model

Use HF_TOKEN secret for inference client

Initial Meta-SRE Agent API deployment