After a job's workspace was deleted, probably by the workspace cleanup thread due to a period of inactivity, polling from Git (without Fast Remote Polling enabled) failed:

No workspace is available, so can't check for updates.
Done. Took 84 ms
No changes

In AbstractProject._poll, if requiresWorkspaceForPolling and there is no workspace and there is a self label, then building is skipped:

if the build is fixed on a node, then attempting a build will do us no good. We should just wait for the slave to come back.

which makes sense if the workspace is unavailable because the slave is offline, but not if the slave is online and the workspace has simply been deleted.