use a fork to fix compatibility with ollama + openweb ui endpoints
update the minuet-ai config to use the updated prompt string and
response parsing customization capabilities that this fork brings in,
which is necessary to use the non-legacy api completions endpoint
this could be lowered if n_completions = 1. this basically causes
minuet-ai to internally do a loop for n_completions where each iteration
runs a separate curl request
most fim-capable models use a "template" that will auto-generate the
correct prompt text for the model using the `prompt` and `suffix` values
in the request data blob, if the `suffix` key is found