Optimized saturation operator

[perrygeo]

Resolves #21

The previous saturation operation went something like:

• swap axes to image order
• convert colorspace using skimage
• swap axes to rasterio order
• apply saturation mutliplier
• swap axes to image order
• convert colorspace using skimage
• swap axes to rasterio order

This PR provides a cython rio_color.colorspace module which provides c functions to do the math directly, then some python wrappers to work efficiently against ndarrays.

See #21 for details of testing and benchmarks.

[coveralls]

Coverage remained the same at 100.0% when pulling 92e63fc on colorspaces into 2ea15fe on master.

[perrygeo]

@sgillies could I get your 👀 on the Cython code in colorspace.pyx? Don't worry about checking the math, I'd just like to get some general feedback on the approach.

• private cdef funcs that return C structs and do the actual colorspace math
• python functions to wrap the private cdef funcs and return tuples
• np.ndarray-typed cpdef functions using loops to apply colorspace conversions to arrays, one of which does the rgb->lch->rgb round trip with saturation adjustment.

Ping me if you see anything that we could do more safely, concisely or efficiently.

[perrygeo]

I couldn't find any advice or benchmarks on the best way to allocate an array of equal shape. This works but there may be more efficient ways.

[coveralls]

Coverage remained the same at 100.0% when pulling 33578a0 on colorspaces into e1dfb37 on master.

[coveralls]

Coverage remained the same at 100.0% when pulling 3937546 on colorspaces into e1dfb37 on master.

[coveralls]

Coverage remained the same at 100.0% when pulling cdfe44d on colorspaces into e1dfb37 on master.

[perrygeo]

@sgillies typing the loop variables didn't do much for speed but changing the order did.

I looked into memoryviews. The primary benefit seems to be writing generic functions that can use cython, numpy or C arrays. I tried memoryviews and it works but there's no speed difference and I don't see the benefit - dealing with numpy arrays explicitly seems like a cleaner approach.

[coveralls]

Coverage remained the same at 100.0% when pulling 5e4a01d on colorspaces into e1dfb37 on master.