REC: polygon area metrics overcount when polygons overlap

[davemfish]

For predictor methods polygon_area_coverage and polygon_percent_coverage, the predictor polygons are intersected with the AOI polygons and the area of the predictor intersection is summed. If the predictor layer includes overlapping polygons, then the area of the overlap is counted once for each polygon. This results in incorrect computed areas, which only really become noticeable when the value is greater than the area of the AOI cell (or percentages over 100).

I can't think of any reason why overlapping polygons should be double-counted like this. I think both of these metrics imply that it is the union of the predictor polygons that matters in terms of assessing its spatial coverage. So we should probably do that union before computing areas.