Address flakiness in commit LSN map test

[morgando]

This test has been flake-y since it uses a sleep to wait for replicants to finish recovery before doing validations on them. The sleep has sometimes not been sufficient.

The changes in this PR just do validations on the master node. Since we're guaranteed that the master is done with recovery by the time the call to truncate returns there's no more need for client-end timing logic.

Alternatively a better waiting method could be used to detect that the replicants are finished with recovery before performing validations on them.