bootstrapping new node after upgrading cluster causes counter columns to randomly have incorrect values

Running the latest version of upgrade_through_versions_test will randomly fail at a rate of about 1 out of 5 runs due to an incorrect counter value. A slightly modified version of the test is attached for reference. This version has trunk eliminated from the versions list and an extra debugging statement.

The problem occurs after upgrading to the 2.0 branch from the 1.2 branch and after boostrapping a new node to the cluster. Best way to repro is just run the test in a loop 5 to 10 times.

Be sure to set PRINT_DEBUG env variable to true and run test with --nocapture to see the debug output. Logs are also included.