Assigning a protocol for an experiment's datasets doesn't always assign the datasets properly
This is an issue with the user/user/fisherface/1/atnt experiment. After pulling it from a local BEAT instance, it using the atnt/3 database. Because the current version of beat.cmdline only supports atnt/5, I pulled that db tried to update all the experiment's datasets by selecting the idiap_test_eyepos (atnt/5) protocol option. It flipped the dev/test datasets when assigning
