All parameters about SLA, i.e. computer-guid & instance-guid, are removed in
favor of a new "sla-dict" parameter, which is easier to implement and much more
versatile.

All changes in the request parameters are incompatible. The old ones are
ignored without warning/error.

For compatibility, the reference of ZEO partition is still "zodb".

Default settings were also fine. Default name of mount-point and FileStorage
file is reverted to 'root' instead of 'main'.

In order not to conflict with a future integration of NEO in ERP5:
- the input schema has a new parameter for external storages.
- zodb-software-type & zodb is used only for internal storages
  and only ZEO is supported.

NEO logging is also enabled for clients.

Implemented using libuserhosts. Only provided to zope processes so far, may be
extended in the future with each process seeing different sets of aliases,
with potentially different resolutions.

Move haproxy httpchk configuration per-listen as zope's webdav returns 401
on (at least) GET / . TCP-level checks are still active, which should be
sufficient.
Also, change zope_family_address_list check to getitem as it's more natural
than concatenating.
Also, stop using apache.zope.backend recipe to add support for per-port
scheme so automatic ERP5Site creation knows which URL is really HTTP.
Also, stop using certificate_authority recipe and call openssl directly
when generating self-signed cert for https/webdavs support.
Also, rework balancer parameters a bit (still undocumented).
Also, make instance-balancer always publish ipv6 value (empty when no ipv6
available), so instance-erp5 always succeeds.

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

Frontend support is desired, but current option set is too weak to cover
even simple cases.

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

Only able to use a pre-existing NEO cluster.