ci: find ~8-9 GiB more stuff to remove

[iliana]

Yet another full disk error prompted this, but this is really scraping the bottom of the barrel for what we can easily remove from the runners at runtime.

This probably won't be the last time we need to worry about this. Some janky ideas for when we inevitably hit the brick wall:

• Move these jobs onto Buildomat-managed Linux VMs. Not sure if this is actually a good idea, but then we're at least controlling how much disk these hosts get.
• Set up a mountpoint on a ZFS or Btrfs disk backed by a file with compression enabled.
• Try to figure out how to reduce the size of the intermediate build artifacts.

Also this updates us to ubuntu-20.04 because I thought it'd be nice to do.

Closes #1380 (again).

[jclulow]

FWIW, I added basic Linux target support to buildomat last week. There's an ubuntu-21.10 target now, that runs in AWS with the same machine specifications as the helios-* targets; these have 200GB disks at the moment.

I have made a rough sketch of a start for the conversion here: https://github.com/oxidecomputer/omicron/compare/buildomat-linux

The Linux test run was a success: https://github.com/oxidecomputer/omicron/runs/7314656250

[iliana]

Backed out the ubuntu-20.04 change so I can merge this.

[iliana]

No longer necessary (#1518)