Make UnsignedBytes.lexicographicalComparator() use Arrays.compareUnsigned when it's available.

[copybara-service]

Make UnsignedBytes.lexicographicalComparator() use Arrays.compareUnsigned when it's available.

This provides another alternative to using Unsafe.

And port the benchmark to JMH, which for now means making it Google-internal. (Not that we have our Caliper benchmarks actually running externally, either, IIRC.)

The benchmarks suggest that the old, Unsafe-based implementation is faster up to 64 elements or so but that it's a matter of only about a nanosecond. The new implementation pulls ahead for larger arrays, including an advantage of about 5-10 ns for 1024 elements.

For now, I've included this implementation only in the JRE flavor of Guava. We could include it in the Android flavor, too, to see if it helps under API Level 33+. But we really would want to do yet more benchmarking for that.

RELNOTES=n/a