Configure default parameters for models

[fmaclen]

Also stuff like "use mmap" or thread count Originally posted by @binarynoise in #279

There's many ways we can go about this:

1. Apply model settings globally. This will mean that all of the models from any server will default to the user-defined parameters. This is how OpenWebUI handles it AFAIK.
2. Apply the settings per server/connection. Means that the models from that server will default to the user-defined parameters.
3. Apply the settings per model. User-defined settings will only be applied when a particular model is used, all others will use the provider's (Ollama/OpenAI) defaults.

Without having done supe deep research, option 1 is probably the easiest implementation of such a feature.

@binarynoise do you have any insights on how this would play out in your workflow?

[binarynoise]

I would prefer option 1, as this would allow using default settings also when using a private tab, my laptop/phone, or when a friend uses my hollama from their PC.

Option 2 would be okayish, because we would lose the benefit of defaults for new clients but would need only one setup per client.

I don't like option 3 because when switching models around a lot, I'd need to apply the same settings over and over. I would only add this as an additional layer.