Lower autoscaling alarm threshold to 50%

We see in the CI, that sometimes the CPU metric doesn't reach
above 60%, so the stack never scales up and the CI fails.
I've never seen it stop below 50%, so this threshold would always
be reached in the past.

The core issue of such a low values are possibly caused
by the length of the load we create in the autoscaled instances.
The load is created only for 2 minutes, which is too low
when ceilometer polls it only every 2 minutes. With bad
luck in timing we will get these low metrics. I'll also increase
the load length in tempest.

ci/vars-autoscaling-tempest.yml

@@ -16,7 +16,7 @@ cifmw_tempest_tempestconf_config:
       telemetry.prometheus_service_url "https://metric-storage-prometheus.openstack.svc.cluster.local:9090"
       telemetry.ceilometer_polling_interval 120
       telemetry.prometheus_scrape_interval 30
-      telemetry.alarm_threshold 60000000000
+      telemetry.alarm_threshold 50000000000
 cifmw_test_operator_tempest_include_list: |
   telemetry_tempest_plugin.scenario
   telemetry_tempest_plugin.aodh