optimize mark_duplicates_and_sort

[chrisamiller]

The tools/mark_duplicates_and_sort.wdl is a bottleneck, especially for WGS. It's expensive, and that's partially because it is long-running and gets preempted. Do some local testing on the cluster to explore options for optimizing it:

• right now the sort and markdup get 8 cores each. Is that the optimal ratio? If one is faster than the other, there'll be wasted cycles.
• If we increase the number of overall cores, how does that affect runtime? (do we saturate I/O? is that different between HDD/SSD?)
• Can we prevent localization of the input files to save an hour or so?
• would giving more ram to the sort part of that step allow it to do less slow writes of temp files to disk and speed things up?

[chrisamiller]

Breakdown of costs/timing for WGS:

sample	seconds	cpuCost	memCost	diskCost	diskType	totalCost
HCC1395 normal	24870.788	0.771547112	0.42860658	0.221073671	local-disk 576 HDD	1.421227363
HCC1395 tumor	46089.762	1.429806839	1.143282152	0.750381156	local-disk 1055 HDD	3.323470147

Both ran on a custom-16-97280 instance