Ignore previous instance work requests in idle_time calculations

This should stop the provisioner from immediately killing workers for being idle, seconds after launching them.

Part of: #1203

Merge request reports

Loading