New issue 549: Projections should switch off parallel reduction method https://bitbucket.org/yt_analysis/yt/issue/549/projections-should-switch-off...
For those simulation types where the global mesh is known, we should use the (slightly
more scalable) global mesh information to pre-construct the quadtree which is then reduced
as an array across all processors. For those simulation types where we do not know the
global mesh information, we should fall back on the quadtree reduction method we used
before, which is in the routine
This will require a flag or some type of discriminator between global/local mesh simulations.