I discovered this while working on JDK-8351409.

The clipping range method in the MinMaxVector JMH benchmark has an overflow issue when calculating `highestInt` and `highestLong`. If they overflow, the range parameter won't work as expected and therefore the branch taken/not-taken expectations won't be matched. When this happens, the results of the benchmark cannot be trusted.