Support custom models

[marcorosa]

Support models hosted in non-AI Core platforms:

• Azure AI Foundry
• AWS Bedrock
• generic vllm servers (like what we have now for mistral -> generalize it)
• Custom AI Core serving executables (e.g., Ollama or vllm via AI Core)