Fully supports check-mode as well, so can be used to check what the
module will do.

Module will delete routes not specified in the config, as well as
updating/adding others, so it brings routing configuration under full
management

Otherwise module will fail to run in check-mode. Maybe this is a bad
idea, however it seemed to me to be fairly harmless in that usually
adding a package is relatively low-impact, thus it could perhaps be
forgiven in this context?

Previous structure didn't work when the key, previously the NID's IP
address, is itself a variable, eg:

 lustre_lnet_lnetctl_networks:
   'o2ib2':
     "{{ server_network_ib_address }}":
       interfaces:
          - 'ib0'
       tunables:

This doesn't appear to work, so instead, we change the structure to be a list
of NIDs:

 lustre_lnet_lnetctl_networks:
   'o2ib2':
     - ip: "{{ server_network_ib_address }}"
       interfaces:
          - 'ib0'
       tunables:

This is being removed as in practice it doesn't work. The problem is
that you can't just restart LNET on a server since the modules will be
in use by any active targets that are mounted.

This means that we can't restart LNET on configuration change - that
kind of action will have to be a separate on-shot playbook run, or by
running the role with state: absent to shutdown Lustre fully, and then
state: present to start it up again fully.

We add:

* a new 'ldisfs_filesystem' module, based off the standard filesystem
module, for use with idempotent formatting of ldiskfs lustre targets

* optional ldiskfs formatting tasks

* optional target mounting tasks

Includes:
- ko2iblnd
- lnet module config
- lnetctl-style yaml config file (subset of options for now)