Scaffold a new inference spec (and evals.py) so you can start versioning.

Creates specs/<name>.yaml from a starter template and, if it does not exist yet,
an evals.py with a couple of example evaluators. Pass a name to choose the spec
(the default is 'summarization'); the spec's name and the file are kept in sync.
Edit the spec to describe your prompt, model, sampling, and evaluation, then run
'dow commit' to capture v1. This only writes starter files - there is no staging
area, index, or refs, and it refuses to overwrite an existing spec.

@examples
dow init
dow init chatbot
