This is a hack to avoid Emulab and systemd conflicts.  Basically, sometimes
our MFSes add swap devices to /etc/fstab that are incorrect (i.e., /dev/hda
instead of /dev/sda).  We could always add 'noauto' to the mount options so
systemd would ignore it, but we have legacy or deployed MFSes to deal with.
When systemd encounters one of these, it tries to start the device and halts
the boot process for a long time while waiting for the device to "start".

We can't edit the bogus /etc/fstab entry before system reads it, because
system reads it while the root is still mounted read-only, very early
on in the startup process.

systemd doesn't allow us to remove units (a swap device is a unit, just like
a service is a unit), so all we can do is, before the swap unit runs, cancel
any pending systemd jobs before they try to initialize the the bogus device(s).