Speed up matrix operations with parallelization

[willtownes]

Suggested by Aaron Lun in kstreet13/scry#7

There's probably some low-hanging fruit with BiocParallelization and DelayedArraying, especially if you can make use of the parallelized crossprod provided by DelayedArray (via the Matrix generic).