Add Ollama to bicep

Certain connectors using local LLMs like Ollama, Hugging Face and LG use the Ollama server when deploying to the cloud. However, the current bicep template doesn't cover this scenario.

Add another ACA instance to deploy Ollama server with the GPU plan.

Reference: https://techcommunity.microsoft.com/blog/appsonazureblog/open-ais-gpt-oss-models-on-azure-container-apps-serverless-gpus/4440836