Provider implementation for HuggingFace's own inference router / API.
Routes through https://router.huggingface.co when using an HF token,
or directly to https://api-inference.huggingface.co for certain tasks.
Handles response normalisation for all 20+ HF inference tasks.