Attempts to access ZFS ctldir via libgfapi can cause gluster
volume mount to freeze. We should prevent this variant of
foot-shooting by adding robust checks to ensure that only
directories within ZFS datasets are used. Since we already
have filesystem.stat output as part of brick validation we
can easily expand existence check to also cover things like
fstype, whether the brick path is empty, dataset mount options,
etc.

This PR makes several critical changes to how the ctdb nodes
file (private IPs) are managed in middleware. Some types of
issues that are addressed by this PR are:

* Inconsistencies in nodes configuration file on different
  cluster nodes
* Duplicate nodes entries being added via ctdb.shared.volume.create
  (e.g. same node twice with different IPs)
* Lack of clear mapping between ctdb nodes and gluster peers

Originally there was easy mapping of entries in the ctdb nodes
file back to the originating gluster peer. This PR changes the
nodes files entry from form of
```
<ipaddress>
<ipaddress>
<ipaddress>
```

to
```
<ipaddress>#{<peer information>}
<ipaddress>#{<peer information>}
```

Above required a change to nodes file parsing function in ctdbd,
which means this PR may not be backported into stable/bluefin.

Currently only peer UUID is stored in the file, but in future
we can potentially expand to include additional information.
Store gluster node UUID in ctdb nodes file. This allows us to
more tightly couple gluster TSP configuration with our CTDB nodes
configuration (e.g. include peer UUID in ctdb.private.ips.query
return).

Since this change requires that the backend maintain the mapping
between nodes and peers, the ctdb private ips API was changed to
require caller to provide a gluster peer UUID for the nodes entry.

The ctdb.shared.volume.create method originally would automatically
append nodes file entries for missing gluster peers based on DNS
lookup results, but this has over time proven to be somewhat less
than reliable (users may have DNS misconfigurations that result
in multiple nodes files entries or incorrect interfaces being used).

This PR allows caller of ctdb.shared.volume.create to optionally
include a list of private address + gluster peer UUID mappings
to be added (if necessary) to the nodes file. This method will now
explicitly fail in the following situations:

* gluster peers that are not present in resulting file
* peer UUIDs in payload that do not exist
* entries in nodes file that do not map back to gluster UUID

This has somewhat wide-ranging impact for how our backend APIs are
used. For instance, if a new gluster peer is added to an existing
TSP, then a private IP for the nodes file must also be supplied.

A new cluster management service is also being added via this
pull request to enforce consistency in how SCALE clusters are
configured (and ensure that proper validation takes place to prevent
issue reports on drastically misconfigured clutsers). It also
provides a stub of an API for adding new cluster nodes (expansion).

The end-goal of this API is to force the caller to provide us with
a payload containing gluster peer, brick, and private address (nodes
file) information for each proposed cluster node.

This reverts commit 23623d5e.

This reverts commit c91cfc25.

This reverts commit 9fc830a9.

This reverts commit ec9ca6f6.

This reverts commit 7c5c0011.

This reverts commit 72e63984.

This reverts commit 1c4d1320.

These are already set in snapshot for CI

* Caleb thinks current failures might be a race condition.  Trying to increase timeout as he suggested to see if it resolves test issue.

* use global job timeout and bump to 120 seconds

* missing f in f-string

* allow_duplicate_serials=True

* add init_gluster to cluster api pipeline

* finalize setting up gluster volume

* add framework for calling pytest-3

* simplify pytest command generation