If there are reads that are scheduled via ExecutorFuture but not run, it's possible with the mirror_reads.js loose test assertions that the test continues. The test would eventually fail at `checkReadsMirroringRate`, where resolved is compared with seen. We can fix this by 1) changing where the metrics are incremented, 2) making the assertions in `waitForReadsToResolveOnPrimary` more strict, or 3) change assertions around `checkReadsMirroringRate`