Switch Ollama distro to use ollama embeddings

[ashwinb]

🚀 Describe the new functionality needed

We should not need sentence transformers for calculating embeddings if we are using ollama. No reason to have a torch dependency. Specifically, we need to update templates/ollama/run.yaml to point the embedding model to the ollama inference provider.

💡 Why is this needed? What if we don't build it?

Dependency on the transformers package is a two-edged sword due to its complexity.

Other thoughts

Need to ensure that client-sdk tests pass (i.e., the standard all-MiniLM alias we use everywhere "just works" -- if there is a more standard HuggingFace ID for that model, maybe we should use that.)

[manojks1999]

/assign

[manojks1999]

@ashwinb can I work on this

[ashwinb]

Thanks @manojks19999 -- please hop on to the Discord if you need any help.