Test the impact of memory alignment

[dievsky]

We could devise a simple enough benchmark to test the impact of alignment on element-by-element operations (e.g addition).

Take a DoubleArray and generate four F64Arrays with different alignments (starting at element 0, 1, 2, 3). Then benchmark adding each of them to another, fixed F64Array. We should see a spike in throughput where alignments coincide.

(This, however, would only be valid if the garbage collector doesn't move the array around between benchmark iterations and shift the alignment.)