Flaky integration tests

[atbore-phx]

We have switched our tests to Semaphore CI (#1266) and we got a execution time boost (~2x).
We also restored the ci_retry function (#1573) because the integration tests are still flaking.

I'm starting to deploy a simple scenario based on tests series to obtain a closer perimeter and to understand which tests have the higher "flaky probability".

Reiterating the same test on semaphore 100 times (deactivating the ci_retry function),
i get those results: link

[atbore-phx]

Series : S1
KO: 14%
higher errors:
acme_test 5
eureka_test 4

[atbore-phx]

Series: S2 (acme and eureka tests are executed first):
KO: 15%
higher errors:
acme_test 7
eureka_test 4

Acme and eureka tests are still the tests with the higher flaky probability.
The test's order doesn't affect the final results so we "could" exclude an hw resource exhaustion issue.

[atbore-phx]

Series: S3 (increasing acme and eureka timeouts).
KO: 8%
higher errors:
acme_test 2
eureka_test 3

An higher timeout on the most flaky tests seems reduce the error probability (especially for the acme test )

[atbore-phx]

This PR #1599 has the main goal to deflake the integration tests.

[ldez]

Closed by #1599