Support fallback to origin openai endpoint if model mapping not found

[Vigilans]

This would make azure-openai-proxy works to support all mode types, and only those model types supported by azure openai are proxied.