[Feature Request] Support Hugging Face Serverless Inference API

[hvaara]

Overview:
Integrate the HF Serverless Inference API to streamline model inference for models hosted on the Hugging Face Hub.

Why?

• Ease of Use:
  Developers can easily fetch model inferences without the need for specialized hardware or infrastructure management. Once the query is completed, there’s no cleanup required. Additionally, there’s no need for downloads, and because the underlying infrastructure is closely integrated with storage, cold boots are quick. Increased parallelism boosts throughput, resulting in less time spent overall on inference.

• Cost Efficiency:
  Using a serverless model eliminates the overhead of maintaining dedicated infrastructure, or cloud infrastructure. Which, let’s face it, can rack up a huge bill - especially if forgotten about.

• Extendability:
  This integration sets a foundation for supporting additional inference services through HF Inference Providers down the line since both use huggingface_hub.InferenceClient.

[hvaara]

Draft PR incoming