Only submit AOI edge geometries to geotrellis

[jterry64]

Users can submit arbitrarily large AOIs, and these can be very computationally intensive to analyze, even if there aren't that many. Since we're already going to storing precalculated results for grids at different zoom levels, I don't think it makes sense to always recalculate the full AOI area, especially for tree cover loss which remains the same for months. Likewise, we're already dividing up large geometries into 1x1 tiles, where we already might have results calculated for complete tiles.

I think we should follow the same datacube algorithm as the lambda raster analysis, and only compute the edge geometries for AOIs. Then we can merge those results with the precalculated grid table.

The question is whether we should do this here in a lambda, and just submit the edge geometries in our TSV to geotrellis, or if geotrellis should do it itself. Doing it here would keep geotrellis simple instead of having to be aware of the grid dataset and deal with that merge. However, geotrellis also already contains this merging logic, and I imagine this could apply to any calculation of large geometries, not just user AOIs.

@thomas-maschler, thoughts?