Flaky test_restart_fast_sync and test_fast_kill #6746

These tests are going to be flaky after #6714. They were relying on restart having a hardcoded 10s timeout and not waiting for all workers. We should probably also rethink the purpose/implementation of these tests—what is the "fast" aspect meant to actually be testing? Maybe restart is no longer the appropriate way to test this fastness?

cc @hendrikmakait