Add Hugging Face and vLLM hosting options

[Weili Shi (shi-weili)]

Related Issue

N/A

Summary

Expand the model hosting guide with managed Hugging Face Inference Endpoints and self-hosted vLLM options alongside Microsoft Foundry Managed Compute.

Changes

• Add Hugging Face deployment, authentication, model sizing, scale-to-zero, and cold-start guidance.
• Add self-hosted vLLM commands and MagenticLite connection settings for Fara and MagenticBrain.
• Reorganize Foundry Managed Compute as a third hosting option and document HTTP 503 cold-start troubleshooting.

How to Verify

1. Review docs/model-hosting-guide.md and confirm each hosting option includes deployment and MagenticLite connection steps.
2. Confirm the option and troubleshooting anchor links resolve to the corresponding sections.
3. Run git diff --check origin/main...HEAD.

Checklist

• Tests added or updated (if applicable)
• Documentation updated (if needed)
• Verified using the steps above