Report client does not handle keystone errors properly

[Oliver Walsh]

_create_resource_provider is wrapped by safe_connect which results in a None return value if keystone/connect exceptions are raised.

On master the result is a TypeError and the rp isn't cached:
      File "nova/scheduler/client/report.py", line 697, in _ensure_resource_provider
        rps_to_refresh or [created_rp])
      File "nova/compute/provider_tree.py", line 268, in populate_from_iterable
        to_add_by_uuid = {pd['uuid']: pd for pd in provider_dicts}
      File "nova/compute/provider_tree.py", line 268, in <dictcomp>
        to_add_by_uuid = {pd['uuid']: pd for pd in provider_dicts}
    TypeError: 'NoneType' object has no attribute '__getitem__'

However on stable/pike _ensure_resource_provider caches the None return value and never retries the resource provider create. Seen in https://bugs.launchpad.net/tripleo/+bug/1767076

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/566148

[OpenStack Infra (hudson-openstack)]

Change abandoned by Oliver Walsh (<email address hidden>) on branch: master
Review: https://review.openstack.org/566148