Currently a user needs to copy several paths to an eval config from a results directory: checkpoint_path, spect_scaler_path, labelmap_json_path.

But since the results directory created by a run of "vak train" always has the same contents, the only things that really change are the parent directory; the filenames are (usually) the same.

We could instead bake in this knowledge and let a user just specify model_path

This would also make it easier to generate configs as in #366 -- One could say vak make-config --eval --model_path=./path/to/model (or something like that)