Handle nodata masks about rio-hist

[perrygeo]

Thinking through what this will take...

• We can just drop nodata in the reference when we convert to 1D, no info lost there
• We have to keep track of the masked value's index, delete it from the values and counts arrays before cdf. Then after the target values have been interpolated we need to re-introduce the fill value at that index, recreate the array and re-mask by position
• We need to ensure that we're writitng masked rasters with nodata values
• Histogram and cdf plots need to also consider mask
• the hard one colorspace conversions (forward and backward) need to retain the mask. This means that we'll need to save the mask and reapply it at various steps. There's some uncertainty as to how this will work - real data might become a nodata value in a given colorspace so the mask will need to be applied by position, not value.
• This is going to add a bit of complexity to the code. Maybe we need to detect the case automatically and make the logic conditional?