Incorrect discrete 'curl' colormap

Looks like something is off when you specify a discrete normalization with the curl colormap.
The code below will produce 4 plots; the first 3 are as expected but the last one demonstrates the issue. Colors are no longer centered around 0 (white).

import matplotlib.pyplot as plt
import cmocean
import numpy as np

data = np.random.normal(loc=0, size=(10, 10))
norm = mpl.colors.TwoSlopeNorm(0, -2, 2)

plt.imshow(data, cmap='cmo.balance', norm=norm)
plt.colorbar()
plt.title('TwoSlopeNorm, Balance (Correct)')

plt.figure()
plt.imshow(data, cmap='cmo.curl', norm=norm)
plt.colorbar()
plt.title('TwoSlopeNorm, Curl (Correct)')

bounds = np.linspace(-2, 2, 21)
norm = mpl.colors.BoundaryNorm(bounds, 256)

plt.figure()
plt.imshow(data, cmap='cmo.balance', norm=norm)
plt.colorbar()
plt.title('BoundaryNorm, Balance (CORRECT)')

plt.figure()
plt.imshow(data, cmap='cmo.curl', norm=norm)
plt.colorbar()
plt.title('BoundaryNorm, Curl (INCORRECT)');

Start by running the issue's four-plot reproduction with matplotlib, cmocean, numpy, TwoSlopeNorm, and BoundaryNorm. Compare the cmo.balance and cmo.curl results under BoundaryNorm, then inspect the curl colormap setup and relevant normalization behavior; done means discrete curl colors are centered around 0 like the continuous example.