Document verified Foundry deployment settings

[shi-weili]

Related Issue

Summary

Update the Foundry model hosting guide with settings verified from deploying Fara 1.5 and MagenticBrain in Azure AI Foundry and running both models in MagenticLite.

Changes

• Recommend Standard_NC24ads_A100_v4 for testing and typical single-user use instead of a larger A100 SKU.
• Clarify that Azure quota requests should be made under Azure Quotas > Machine learning for Standard NCADSA100v4 Family Cluster Dedicated vCPUs in the same region as the Foundry project.
• Clarify that one NC24 instance consumes 24 dedicated vCPUs, so the usual two-deployment Fara + MagenticBrain setup needs 48 dedicated vCPUs at instance count 1.
• Note that Foundry may default to 3 instances and users should reduce instance count to 1 for testing or typical single-user use.
• Correct the MagenticLite connection instructions to use the Foundry REST endpoint through /v1, the model ID names Fara1.5-9B and MagenticBrain-14B, and each endpoint's primary key from the Consume tab.

How to Verify

1. Open docs/model-hosting-guide.md and review the Foundry prerequisites and deployment table.
2. Confirm the guide points users to Azure Quotas > Machine learning and the Standard NCADSA100v4 Family Cluster Dedicated vCPUs quota family.
3. Confirm the guide explains that two concurrent NC24 deployments need 48 dedicated vCPUs.
4. Confirm the MagenticLite connection table uses Fara1.5-9B for the browser-use model and MagenticBrain-14B for the orchestrator model.

Checklist

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