Use memcached based cache in nova in all devstack-tempest jobs

Using memcached based cache for metadata api to cache data for
instance always after firts metadata request is send can solve
problems with unable to ssh to instance due to not configured
public-key on instance.

In Nova metadata api service data related to one instance should
be cached after it is collected first time. So e.g. CirrOS based
image first sends request to get instance-id and next request is
to get public-key.
If those two requests will be processed by 2 different workers on
Nova's side, it will collect all data for instance twice. And if
this will take more than 10 seconds in case of request to get
public-keys, CirrOS script will fail and not configure public-key
on guest VM.

Using memcached based script will make this cached data available
for each worker so it will be cached always during processing
instance-id request. And this request for instance-id is retried
by CirrOS image in case of failure so this shoudn't cause problems.

Change-Id: I8efedb8c5ae073bd2cb4034369bfcff4b67eb868
Closes-bug: #1836642
Closes-bug: #1808010

.zuul.yaml

@@ -16,6 +16,16 @@
       devstack_services:
         tempest: true
       devstack_local_conf:
+        post-config:
+          # TODO(slaweq): remove this hardcoded options when
+          # https://bugs.launchpad.net/devstack/+bug/1837980
+          # will be fixed and Devstack will support configuration
+          # of cache in Nova
+          $NOVA_CONF:
+            cache:
+              memcache_servers: "localhost:11211"
+              backend: "dogpile.cache.memcached"
+              enabled: "True"
         test-config:
           $TEMPEST_CONFIG:
             compute: