batch job driver pods are OOMKilled before export_workspace finishes

[bossie]

Joris C. has some jobs that seem to be able to write their GeoTIFF assets fine [to the default workspace] before their driver pods are OOMKilled; the assets (of 2GB in size) are not being put into the workspace.

Example job: j-2411068168ad4e0e9596d1df233df0fc

A theory is that ObjectStorageWorkspace is optimized for speed rather than memory; if Python memory is not capped on the driver like it is on the executors (TBC), this might explain why driver pods are OOMKilled.