Use correct Kernel dimensions and optimize. Fix #786

[JimBobSquarePants]

Prerequisites

• I have written a descriptive pull-request title
• I have verified that there are no overlapping pull-requests open
• I have verified that I am following matches the existing coding patterns and practice as demonstrated in the repository. These follow strict Stylecop rules 👮.
• I have provided test coverage for my change (where applicable)

Description

Our current gaussian kernels are too small and fail to capture the full curve leading to boxed output. I've updated the algorithm based on the information found at http://chemaguerra.com/gaussian-filter-radius/ to correct the output.

[antonfirsov]

Haven't gone too deep, but looks good at first look.

We should probably add at least one reference-image based test case to ensure correctness, with a reference image generated by CSS.

[JimBobSquarePants]

We should probably add at least one reference-image based test case to ensure correctness, with a reference image generated by CSS.

@antonfirsov I would but CSS blur over-samples past the edge of the image so we can't compare.

Additionally we use a constrained kernel when sampling, capturing 97% of the Gaussian integral but ignoring changes generally invisible to the human eye when compared to an infinite kernel.

[codecov]

Codecov Report

Merging #789 into master will decrease coverage by 0.03%.
The diff coverage is 95.83%.

@@            Coverage Diff             @@
##           master     #789      +/-   ##
==========================================
- Coverage    88.7%   88.66%   -0.04%     
==========================================
  Files        1012     1012              
  Lines       43248    43222      -26     
  Branches     3127     3120       -7     
==========================================
- Hits        38363    38324      -39     
+ Misses       4181     4160      -21     
- Partials      704      738      +34

Impacted Files	Coverage Δ
...cessing/Processors/Convolution/BoxBlurProcessor.cs	100% <100%> (ø)	⬆️
src/ImageSharp/Primitives/DenseMatrix{T}.cs	86.53% <100%> (+14.79%)	⬆️
...ts/ImageSharp.Tests/Primitives/DenseMatrixTests.cs	100% <100%> (ø)	⬆️
...ng/Processors/Convolution/GaussianBlurProcessor.cs	92.59% <88.88%> (+19.25%)	⬆️
...Processors/Convolution/GaussianSharpenProcessor.cs	93.93% <92.85%> (+14.62%)	⬆️
...xelFormats/PixelOperations{TPixel}.PixelBenders.cs	89.34% <0%> (-9.84%)	⬇️
...Processing/Processors/Transforms/TransformUtils.cs	88.29% <0%> (-5.32%)	⬇️
src/ImageSharp/Common/Helpers/ImageMaths.cs	82.05% <0%> (-5.13%)	⬇️
.../Processing/Processors/Transforms/CropProcessor.cs	96% <0%> (-4%)	⬇️
...ocessing/Processors/Dithering/ErrorDiffuserBase.cs	93.75% <0%> (-3.13%)	⬇️
... and 20 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 590609a...d7a4688. Read the comment docs.