chore(backend): ci test flakiness fixes

[rozzilla]

Fixes applied:

• always define 4 CPU (as by Github docs for standard Linux hosted runners) so that we can always test (either locally and on the CI) with the right amount of CPUs
• use a more reliable way to check for metrics to be populated rather than a simple wait

Also, tests now take half of the time to be executed:

• local results (on my laptop) => from duration_ms of 10263.82 to 5034.62
• CI results => previously 18971.10ms vs now 8374.50ms

[mcollina]

lgtm