Add LiteLLM or llama.cpp support to Generate skills

So far, only models like Anthropic or OpenAI are available for the generation of skills. I noticed that Eval has a way to load custom models and use them for evaluation. However, it doesn't seem possible to use custom models to `generate` skills. I'd like to have support to also add open source models or custom models to generate skills as oppose to proprietary only.

```bash
# Evaluate on local model (llama.cpp server)
upskill generate "parse YAML" \
--eval-model "unsloth/GLM-4.7-Flash-GGUF:Q4_0" \
--eval-base-url http://localhost:8080/v1
```