branch predictability during hot lookup benchmarks is too high

[nbronson]

The hot lookup benchmarks repeatedly look for the same key. This allows the processor's branch predictor to learn the correct pattern of branches even for hash table algorithms that have unpredictable branches, biasing results in their favor. We should construct a sequence of keys for use in all of the inner loops that currently have a limit of kOpsPerKey.

Feel free to assign to me, I plan to fix this.