Also, generate password randomly if not provided instead of using an
hardcoded value.
Publish them as they can be generated internally.

It's already always provided, and looks dangerous: it is mutated during
execution.

Otherwise, it prevents network access to Zope on a freshly created
instance, as site object does not exist.
Leave the parameter undocumented as it is still not organised in a
maintainable way.

It is superseded by per-zope value, published by Zope partitions.

Also, fix frontend request.

Before, a random password was generated for each group of zopes.

Keep it undocumented, as haproxy config in its current state cannot be
sufficient to get properly configured haproxy/apache processes. It needs
a significant rework before it makes sense to document it.

Drop instance-erp5-single and instance-zope:
All deployments of ERP5 instances must be covered by cluster-style
deployment, and tweaked by parameters rather than software type.
Drop instance-varnish:
This is not used currently, so avoid giving the impression that it is
supported. Revive when properly integrated.

This work has nothing to do in erp5-component branch, and is preventing it
from being merged to erp5 banch more often.
Conflicts because of non-reverted
6d1a9805 have been resolved to that its
intent is preserved with non-cluster ERP5 SR.

This reverts commit 77976a14.
This reverts commit 7aa1d92c.
This reverts commit 80062015.
This reverts commit 36ed58f2.
This reverts commit 1fe3f466.
This reverts commit 7731726a.
This reverts commit eb2016c4.

In preparation for better ERP5 clusterisation.

It is not used in xwiki and lamp, does not work out of the box in erp5, and
prevents erp5 from moving further toward IPv6 support.
Do not drop its component SR and recipe just yet, as it might find some
use in the near future.

Kept backward compatibility about "production" software_type.

All ERP5 profile is now in a single place.
Moreover, use slapos.cfg stack.

Also, stop referencing reference implicitly pulled parts.