validate: implement validation strategy for ufunc benchmarks

@weiya711 double check that I got this correct.

When some flag is passed to the benchmarks, both the taco and python benchmarks should dump their result tensor to a given directory with some unique key (that is the same between taco and python). Once all of the benchmarks run, we can run a diff between each of the files (generated by python and taco) to ensure that the taco code is correct.