Allow indexing with Ollama to use any model

[johnnyasantoss]

On the OpenAI compatible setup, Roo allows entering any model and any model dimension. It should be the same with Ollama.

I want to use qwen3:0.6b locally as it's shown (MTEB dash) to be much better than the current alternatives recommended by Roo on the setup page.

[nikhil-swamix]

i raised similar enquiry, i was trying using deepinfra
#5572
@johnnyasantoss let me know what you think.