Updated tests model to "LLama 3.2 1B Instruct"

[Lyrcaxis]

This PR makes the default model used for CI tests be the smaller LLaMA-3.2-1B-Instruct-Q4_0 instead of the significantly larger llama-2-7b-chat.Q3_K_S. This should speed up CI tests and save a couple GB in space.

[martindevans]

Thanks a lot for this, it should really help out with test stability and speed in CI! I've pointed out a couple of minor issues that need fixing