Faster Gaussian blur implementation

[Keavon]

The current implementation of Gaussian Blur runs through each pixel and blurs it with all its neighboring pixels within the kernel radius. That's great for other types of kernels, but the special case of Gaussian Blur can be drastically sped up by blurring just the X axis followed by just the Y axis.

(An even faster approach at larger radius values, using an FFT and IFFT, can be tackled later in another issue.)

[kbujari]

Hi, I'd like to try taking this one on.

[Keavon]

Last time we tried tackling this, it was blocked by some larger architectural issues. But that might be solved by now. @TrueDoctor do you think this is approachable now, or is it too involved for a beginner?

[Keavon]

@kbujari I'm told by @TrueDoctor that this is actually not blocked anymore. Please proceed!