PERF: vectorize shapely_to_path coordinate extraction

[greglucas]

Rather than pulling out each sub-geometry .xy's coordinates, we can do one get_coordinates() call and fill in the codes all at once. Using cumsum to incrementally keep track of which ring we are in for the code placement.

I get about 50-100x speedup in the following example.

import time
import numpy as np, shapely.geometry as sgeom
from cartopy.mpl.path import shapely_to_path

t = np.linspace(0, 2 * np.pi, 50_000)
holes = [np.column_stack([np.cos(t[::100]), np.sin(t[::100])]) * 2 + [x, 0]
            for x in np.linspace(-40, 40, 20)]
poly = sgeom.Polygon(np.column_stack([np.cos(t), np.sin(t)]) * 100, holes)

t0 = time.perf_counter()
shapely_to_path(poly)
t1 = time.perf_counter()
print(f"Time taken: {t1 - t0:.5f} seconds")

[greglucas]

get_coordinates() defaults to 2d arrays, so we can use it directly here even if someone passes a geometry with Z.

https://shapely.readthedocs.io/en/2.1.2/reference/shapely.get_coordinates.html