Refactor visualize cross section

In GitLab by @Huite on Apr 15, 2022, 14:30

Cross section currently builds a QuadMesh. It seems somewhat unlikely that this is faster than a straight up PolyCollection. This is more flexible, and easier to construct, e.g.:

x = da["x"].values
nx = x.size
nlayer = subsoil["layer"].size
x0 = np.broadcast_to(x - 0.5 * dx, (nlayer, nx))
x1 = np.broadcast_to(x + 0.5 * dx, (nlayer, nx))
y1 = (da["thickness"].cumsum("layer") + da["zbase"]).values
y0 = y1 - subsoil["thickness"].values

x0 = x0.ravel()
x1 = x1.ravel()
y0 = y0.ravel()
y1 = y1.ravel()

values = subsoil[var].values.ravel()
nface = values.size
vertices = np.empty((nface, 4, 2))
vertices[:, 0, 0] = x0
vertices[:, 1, 0] = x1
vertices[:, 2, 0] = x1
vertices[:, 3, 0] = x0
vertices[:, 0, 1] = y0
vertices[:, 1, 1] = y0
vertices[:, 2, 1] = y1
vertices[:, 3, 1] = y1

kwargs = {"cmap": "turbo"}
valid = ~np.isnan(values)
vertices = vertices[valid]
collection = PolyCollection(vertices, **kwargs)
collection.set_array(values)
primitive = ax0.add_collection(collection)

The issue does not name a file or test. Start by locating the cross-section visualization that currently constructs a QuadMesh, then compare it with the proposed PolyCollection geometry. Done means the cross section uses the new collection while preserving its rendered values and handling of invalid data.