Add union-find benchmark

[aprokop]

This would allow to evaluate any changes in the union-find data structure.

[aprokop]

Please review.

[dalg24]

Why a dependency in Boost.ProgramOptions and not in a Google benchmark?

[aprokop]

@dalg24

Why a dependency in Boost.ProgramOptions and not in a Google benchmark?

I think the dependency on boost.program_options is required because I want to allow running the benchmark with different problem sizes, which would depend on the backend.

[dalg24]

If we are worried about minimal dependency we can parse the command line arguments ourselves. I find it a little surprising (the not using Google benchmark).

[aprokop]

If we are worried about minimal dependency we can parse the command line arguments ourselves. I find it a little surprising (the not using Google benchmark).

Setting Benchmark seems a bit heavy. We only have a single benchmark currently using it (bvh_driver). It is doable, though.

[Rombur]

I still finds this confusing because if I set the parameter to true, we output no on the screen.

[aprokop]

My perspective is that the default behavior without specifying the flag should be allowing the loops. Then, the boolean switch would be used to prohibit the loops. It is unclear to me how would one use a boolean switch in this context.

[aprokop]

Rebased on master to include #799.