Tests sleep for 1 epoch duration expecting the global epoch to advance

[mbutrovich]

We've found a number of tests (serializable_transaction_test, for example) that sleep a thread for exactly the duration of an epoch and expect the global epoch to have advanced. However, this window may be too short depending on how threads are scheduled on the host, and causes tests to randomly fail. These tests should either sleep a bit longer, or deterministically advance the global epoch.