Not enough disk space on CI Batch runners

[iandees]

When large sources (like USDOT's NAD in openaddresses/openaddresses#7253) come up, the default image used by the managed Batch compute environment doesn't have enough disk space to download and process the data.

To solve this, I tried to do something like 1252532, but that compute environment is unmanaged. I think we might want to convert the t3 compute environment to unmanaged as well, possibly using the same AMI that the mega environment uses?