Improve accuracy of mean_hue for Lab and Luv colors

[kimikage]

This calculates the bisection vector in Cartesian space and then obtains its angle.

This is still in the idea stage and has not been fully studied for accuracy and speed; I would like to prioritize the merging of PR #494.

Accuracy

1000000 samples	mean error	maximum error
Float32 before	6.4e-6 °	4.0e-5 °
Float32 after	4.2e-6 °	2.3e-5 °
Float64 before	12.0e-15 °	7.7e-14 °
Float64 after	7.7e-15 °	4.4e-14 °

Speed

Unfortunately, there is little speed improvement.

julia> a = Lab{Float32}.(rand(RGB{Float32}, 1000, 1000));

julia> b = Lab{Float32}.(rand(RGB{Float32}, 1000, 1000));

julia> @btime colordiff.($a, $b);
  69.342 ms (2 allocations: 3.81 MiB) # before
  67.334 ms (2 allocations: 3.81 MiB) # after

[codecov]

Codecov Report

Merging #505 (5d7ece3) into master (f71d7a1) will increase coverage by 1.54%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #505      +/-   ##
==========================================
+ Coverage   93.84%   95.38%   +1.54%     
==========================================
  Files           9        9              
  Lines        1154     1171      +17     
==========================================
+ Hits         1083     1117      +34     
+ Misses         71       54      -17     

Impacted Files	Coverage Δ
src/differences.jl	99.28% <100.00%> (-0.02%)	⬇️
src/utilities.jl	98.34% <100.00%> (-0.35%)	⬇️
src/conversions.jl	99.38% <0.00%> (+<0.01%)	⬆️
src/colormaps.jl	98.13% <0.00%> (+3.08%)	⬆️
src/algorithms.jl	75.00% <0.00%> (+14.04%)	⬆️

---

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 f71d7a1...5d7ece3. Read the comment docs.