Buffer mask for signed distance by dlat

[xylar]

This should make sure the mask includes the various boundaries.

[xylar]

@mark-petersen, I noticed that the signed distance function had the wrong function for the row containing the North Pole for any shape that extends that far north. This seems to be because the rasterize function is sensitive to round-off errors (or something along those lines). The easiest solution I could find was to expand the mask used to determine the sign by dlat so we can be sure it cover everything we want. This might mean the signed distance isn't that accurate right around zero, so I want you to let me know if you see any side effects of this. I've been using this in #518 and haven't seen any bad results so far.

[mark-petersen]

Looks good. Approved based on @xylar's testing.