feat: automatically adjust default gpu_layers by available GPU memory

[mudler]

Is your feature request related to a problem? Please describe.
Having defaults high number of GPU layers doesn't always work. For instance big models can overfit the card and constrain the user to configure gpu_layers manually

Describe the solution you'd like
With libraries like https://github.com/gpustack/gguf-parser-go we could get along and identify beforeahead how much gpu vram could be used and adjust the default settings

Describe alternatives you've considered
Keep things as is

Additional context