Follow-up: Ensure model response capture during duration collection
This is an MR to fix the phenomenon that is described here: https://gitlab.com/gitlab-org/gitlab/-/issues/416330+
This MR still writes the output of the model evaluation to /results/v1 folders. If any testing of this MR occurs, ensure that nothing in the version 1 folders is overwritten.
Note: This MR also renames avg_duration to average_duration as requested in this comment. It was supposed to be renamed before the first round of data collection, but was forgot about.
Upon merge of this MR, the issue linked above will be closed.
Edited by Dylan Bernardi