BUG: Fix fill_between when NaN values are present

[tonysyu]

When NaN values are passed to fill_between, the nan value gets replaced by a value at the origin. This PR reuses the function's mask handling to produce a more user-friendly result.

Here's an example demonstrating the issue.

import numpy as np
import matplotlib.pyplot as plt

x = np.linspace(0, 2*np.pi, 50)
y = np.sin(x)
y[10] = np.nan
y[30] = np.nan
plt.fill_between(x, y)
plt.show()

[efiring]

I think a simpler and more obvious way to accomplish this is change np.asanyarray to np.ma.masked_invalid in the three lines near the top of the function. There is also a cbook.safe_masked_invalid, but it has no docstring, and I don't know what danger it is protecting against--even though I see that I was the one who added it, back in 2009. I suspect it might have been a problem with earlier numpy versions, because a little quick checking does not turn up anything that ma.masked_invalid can't handle.

[tonysyu]

@efiring Indeed, np.ma.masked_invalid is much cleaner. Updated PR.