* Allow RDMA capable interfaces to be exposed for regular networking

When JBOF support was added, we added *all* RDMA interfaces to
rdma.interface.internal_interfaces.  Instead only add those currently
in use for JBOF purposes.

Likewise, in rdma.get_link_choices omit any interfaces that have been
configured for general networking.  Uses interface.get_configured_interfaces

In certain scenarios parameter combinations provided by API
consumers could cause spurious validation errors on mismatch
between acltype and aclmode.

If for some reason a process has an open handle on our old
system dataset, log an error message and proceed with a lazy
umount to avoid leaving system in broken state.

We weren't raising a validation error on empty string for SMB
share name.

Typo in kwarg passed to CrontabSchedule.

When moving the system dataset to a new pool we first temporarily
mount the system dataset in the middleware run directory before
running rsync to copy all of the data to the new path. During
a recent refactor of the system dataset heirarchy the nuance of
the temporary mount point was lost.

This commit does the following:

1. Fixes a bug whereby query for processes owned by paths would not
   list the paths within the `paths` variable in output
2. Adds ability to pass a parameter to include files opened by
   the middlewared process in the output to assist in troubleshooting
   issues where the middlewared process is what is preventing system
   dataset umount from occuring.
3. Passes new parameter when handling errors during systemdataset
   setup.
4. Parses the recursive umount error message for the dataset that failed
   and pass that to the call to look up processes.

* remove unnecessary keys from webui.enclosure.dashboard

* dont convert rotation rate to str

This commit improves performance of the audit.query endpoint in the following ways:
1. Majority of combinations of query-filters and query-options are now passed directly
    to auditbackend.query where they are converted into SQL. This avoids having
    to manipulate potentially very large lists within filter_list.
2. Parse user-provided `query-filters` to look for opportunities to limit the
    number of audit databases we query.
3. When multiple databases are the target of a particular audit.query call,
    perform the queries concurrently via asyncio.gather.

Do not change `readonly` property when we are running in the installer, and it was not set yet (#13274)

* Increase threads_num and turn off volthreading for iSCSI ZVOL extents

* Add pool.post_import hook to turn off volthreading if necessary

Remove the legacy_validate method that was formerly used by the
legacy django middleware for validation. Simplify check for
whether to update our stored machine account keytab.

* Make routines coroutine compatible.

* add JbofModels enum

* add get_jbof_slot_mappings_function

* improve redfish client cache_get() method

* update callers of cache_get()

* update supporting methods to map jbofs

* add jbof_enclosures.py

* fix import

* fix issues round 1

* fix issues round 2

* fix CI

* remove unnecessary redfish.cache_get call

* Mark sensitive information in crypto plugin as private

* Fix OROperator usage

* Add integration tests

* Improve netdata setup

* Add spec for system dataset's children

* fix mount dataset bug

* Add check for partition start and exclude those that are less than 30MiB from the start and end of the drive

These tests fail because we are enterprise-licensed.

Our HA vms are always enterprise-licensed and so we should bypass
tests related to license in the AD plugin.

Use IP address for controller1 to perform initial configuration.

Having our middleware client hit a socket timeout usually means that
the test VM is seriously broken. We should swiftly abort test run
in this case.

This API call needs to be made via the ws_client fixture rather
than normal middleware call test asset.

* simplify to 1 nvme function

* simplify and clean-up jbof/crud.py

A recent change (dade5605) disallowed a hostname from being supplied
as the ip parameter to runtest.py.  Work around this new restriction.
(Only useful in limited circumstances e.g. running tests locally)

* Added a 'quick' method to generate a single drive partition table
Fixed bug (introduced with Electric Eel) where we failed to update the kernel partition table
Added CI test to prevent regression on the partition table update.

* Changed the method by which we request the partition table update.  We now use the `BLKRRPART` ioctl call from `retaste.py`

If networking is not fully configured, kinit attempt may fail with
EAGAIN. Retry several times before giving up. Since the actual kinit
has a timeout set for 30 seconds, this loop may end up being quite
long-running and so convert directoryservices.initialize into a
middleware job. The only direct caller for this private endpoint
is smb.configure which is itself a job and so we don't have to worry
about any potential timeouts here.

* add zpool disk read/write/chksum errors

* update docstring in webui.enclosure.dashboard

* standardize key placement