This is mostly because the Technical Committee decided we should stop
supporting non-merged-usr, but also because we *did* find
issues *without* merged-usr. In any case, our old configuration will
stop being supported in bookworm, so let's stop accumulating legacy
here.

We attach one minio node per gitlab pages host. This might not be how
it works, but it seems simpler to show it that way: worst case, the
minio host gets decoupled and the gitlab pages webserver caches its
content anyways.

This reverts commit fdb4264d and
commit b8483506.

This is so obviously wrong, I don't even know where to begin. Plus it
still doesn't unbreak GitLab. Ugh.

The graphic is now really out of date and not really useful anymore:
it shows that we are basically finished, but that we're also dragging
old machines around for quite a while.

We need to build a new dataset for bullseye now, and it doesn't make
much sense to keep all that old data around anyways.

it's just one possible solution on the deliverability issue, which
will probably require a TPA-RFC to officialize

This was discussed during today's checkin and mostly changed to
accomodate people on the west coast, especially now that we don't have
people on the team in Europe checking in on a regular basis during meetings.

We are using "Must have" instead of "Need to have" there, and it seems
like a more natural wording that we should also use here. It also
happens to reuse RFC-like vocabulary so it may be more explicit for
some people.