Use more accuracy when calculating variance. Fix #866

[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

Use double precision to calculate variance. This prevents early escape from color count calculation. Fix #866

Also cleaned up a little.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@cb18952). Click here to learn what that means.
The diff coverage is 88.92%.

@@            Coverage Diff            @@
##             master     #874   +/-   ##
=========================================
  Coverage          ?   88.96%           
=========================================
  Files             ?     1018           
  Lines             ?    44578           
  Branches          ?     3235           
=========================================
  Hits              ?    39661           
  Misses            ?     4196           
  Partials          ?      721

Impacted Files	Coverage Δ
tests/ImageSharp.Tests/TestImages.cs	100% <ø> (ø)
...ssing/Processors/Quantization/QuantizeProcessor.cs	0% <0%> (ø)
.../ImageSharp.Tests/Quantization/WuQuantizerTests.cs	100% <100%> (ø)
...cessors/Quantization/FrameQuantizerBase{TPixel}.cs	98% <100%> (ø)
src/ImageSharp/Formats/Png/PngEncoderCore.cs	93.53% <100%> (ø)
src/ImageSharp/Formats/Gif/GifEncoderCore.cs	93.9% <75%> (ø)
...rocessors/Quantization/WuFrameQuantizer{TPixel}.cs	91.86% <85.02%> (ø)

---

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 cb18952...b42d93e. Read the comment docs.

[antonfirsov]

Can we introduce a test that fails with the old code and is passing using the updated one? Most straightforward: PNG test based on the image in #866. (Not sure if it's the best.)

[JimBobSquarePants]

I guess... means writing a histogram implementation though a reference image might be subject to change as the quantisation+encoding process is almost as much about aesthetics as anything.

I've added the unit tests from the original implementation.

[antonfirsov]

I will check again as soon as I get home.

[antonfirsov]

@JimBobSquarePants I re-executed the tests after replacing the WuFrameQuantizer<T> implementation with the code before the change, and all tests passed. This means that the improvement introduced by this PR is not covered!

I think there is an easy-to-code, implementation-agnostic approach to test IQuantizer implementations:

1. Create an Image<T> from QuantizedFrame<T> representing the quantization result
2. Compare the quantized image against the original one using a tolerance

With this approach the old code should lead to a test failure using the image from #866!

[antonfirsov]

2 colors before VS 48 colors after. Looks pretty good. Great that the you also added the QuantizedFrame<T> refactor. Let's merge this in!