Add functionality to allow LLM to choose appropriate input files for script context

[bzorn]

Instead of a user writing

def("INFO", env.files, ...)

the user would say:

modeldef("INFO", "NL describing criteria", env.files) and the model would determine which of the files in the current folder/repo would be useful in the query.

Or maybe this would just be implicit in the NL description (as a header parameter like "model-chooses-content: True"