Docs about using MLP outputs (vs CRF)

[ebgoldstein]

In some instances, the MLP outputs are better than those processed with the CRF:
see for example: #42 (comment)

In these instances, an operator should run use utils/plot_label_generation.py to get the MLP output, and use this instead of the CRF.

there could be documentation on this workflow.

optionally, Doodler could save both outputs, and then rely on some sort of QA/QC procedure to pick the better output..

Sniffer ( https://github.com/2320sharon/Sniffer & https://github.com/2320sharon/Streamlit_Sniffer) could have a role to play here