The OSX machine us very slow on CI and exhibited a couple additional
flakiness in the test:

• The block generation in the wait_until loop might be repeated so many times that halving occurs and cause the latter test to fail to reach a quorum due to not enough stakes.
• The improvements from D11834 were not enough to guarantee a 100% success rate on super slow machines, so we are using a double loop to wait for the expected event and fix this issue.

On OSX and Linux:

./test/functional/test_runner.py abc_p2p_getavaaddr