optimize pixel selection

[alee]

Currently there are a few "degenerate" study area selections that cause problems with our naive minimal bounding box algorithm that selects the relevant pixels from the dataset geotiffs

• a thin diagonal strip that intersects many pixels but with a small overall $km^2$ area will timeout
• multipolygon study areas
• ?

See if there's any GIS wizardry we can do to support these kinds of queries. #39 may also help if we can identify these cases early on and move to an asynchronous model of execution, e.g.,

1. user makes a request that the system determines will take a while
2. system returns a url status page that will eventually report the requested results

related issue and comment from @bocinsky : openskope/skopeui#196 (comment)