reshape default mem for depletion tasks

[dpark01]

Until we replace Picard FilterSamReads upstream, for the short term, we should increase default mem allocation to depletion steps.

Currently, the deplete_taxa and filter_to_taxon tasks are allocated 8 cores & 15GB CPU (both GCE & DNAnexus). May want to consider 8 core 32GB (mem2_ssd1_v2_x8 / n1-highmem-8).

On the kraken-side, the filter_bam_to_taxa task is currently 2-core 7GB but we may want to bump to 2-core 13GB (n1-highmem-2 / mem3_ssd1_v2_x2)

[dpark01]

Actually, filter_bam_to_taxa probably doesn't need more RAM since it's likely giving enough to Picard, but deplete_taxa is only giving 50% RAM to the JVM. Maybe we should just bump that fraction a bit