Allocate reasonable memory for counts step on computing clusters

[Nick-Eagles]

SGE, SLURM, and JHPCE configuration requests 140GB of memory for the CountObjects process. While an amount of memory this high has been needed before for large experiments, 140GB is typically unreasonable and excessive. The best solution for this is likely to use a closure to scale the memory request by number of samples (dynamically), potentially with a hard maximum. This may require some testing to determine an appropriate approximate function for memory vs. number of samples.

[lcolladotor]

We could start by having a more reasonable default though, say 50GB. Then even if we request more memory than we need to, we might waste less of it.