Upolygon

Introduces `rle_encode` and `rle_decode`.
Useful for when interacting with the coco format.

Roughly 15 times faster than solutions written in pure python.

draw_polygon` sometimes segfaulted if a path had negative x,y coordinates or at lied exactly at the mask border.

New functionality to simplify polygons, automatically applied lossless to find_contours.

Find_contours: single white pixels could cause an infinite loop due to two uninitialized variables in a code path. This is now fixed with a additional test case.

draw_polygon` should be more or less feature complete.

On smaller polygons (< 150 points, 1000x1000 mask) with few intersecting edges, 1 call takes ~ 0.3 ms (opencv 0.1-0.2 ms), a tad slower.
On larger polygons (500 points, 1000x1000 mask) with many intersections , 1 call takes ~ 7 ms (opencv ~ 68ms), much faster.

`find_contours` is working but the output is very point dense, even on straight lines.

Feedback is welcomed.