The idea of this patch is to provide a more comprehensive baseline for the optimization work I'm undertaking.

Summary:

• Benchmark take when indices are monotonic and contain no nulls. Monotonic takes perform much faster because it accesses memory consecutively rather than at random
• Test null percentages down to 0.01% (1% is even a lot of nulls, and obscures behavior between 1% and 0%).
• Benchmark indices/filter-mask with and without nulls, because there may be faster code paths for the no-nulls case
• Benchmark when values being taken/filtered are all not null
• Benchmark filtering/taking smaller strings. The benchmarks were doing strings of size 0 to 128 -- realistic workloads generally will be working with smaller strings, so I set a range instead of 0 to 32 with 16 the average