Preload the model into Ollama to get faster response times

[wintermeyer]

When Ollama has to load the model it takes longer for it to answer our requests. We can force it to preload the model by a simple empty request:

curl http://localhost:11434/api/generate -d '{"model": "llava:13b"}'

We already know what pages are likely to ask the LLM for help. When those pages are accessed by the user we should preload the model with a spawn (so that we don't shoot ourself in the foot by this).

There is little harm by doing this when the model is already loaded but a big improvement when it isn't.