Benchmarking grouped matrix averaging methods

[bschilder]

Wanted to flag this issue here bc I realize there's quite a bit of overlap in many of the these functions between packages.

https://github.com/PeteHaitch/DelayedMatrixStats/issues/82

Update: I've been noticing a higher than expected number of zeros when computing grouped means on sparse DelayedArrays (relative to other similar datasets that were processed using dense base-type matrices). Wondering if this might be related to the point precision.