Instead of mucking around with the awful iscsi target names we're
stuck with, just go ahead and rescan all the LUNs. There doesn't seem to
be a difference in the delay between rescanning a signal target versus
all targets (almost none).

Also, the previous version failed to include the instructions to
rescan/resize on *all* the cluster nodes, because that doesn't happen
automatically and if the instance is happens to be restarted on another
node, it could fail to boot or worse, cause io errors.

This actually impacts a lot of things, as we need to take those
servers into account when we do SPF and DKIM. Right now, I think the
plan is to grant those servers their own DKIM keys.

Also note the confusion about the name of `relay` which I thought I
already did cover, but now that becomes a little more obvious as we
start to refer to those things.

it's what we used in TPA-RFC-15 and seems to work.

see team#40363

The idea here is that we don't need to justify every step in the
"actual changes" section. The section will be easier to read that
way. Regrouping this in the lengthy "other alternatives" section will
make the text lighter.