Accelerate metabat with GPU

[Rob Egan]

There are a few stages of metabat that could greatly benefit by using a GPU to process the data.

The N^2 matrix is already optimized to perform its calculations by tiles and these tiles can be offloaded to a GPU fairly easily.

Additionally TNF calculations can likely be accelerated.

Unlikely that LPA clustering can be accelerated, but MCA likely could

[Rob Egan]

Attempted and failed. This would really require a refactoring of the use of std::string and other helpers and also the removal and replacement of the OpenMP optimizations. Not worth the effort at this time.