Improve speed and mean accuracy of pow12_5

[kimikage]

pow12_5 is used in the RGB{Float64}-->XYZ conversion.
This ensures sufficient accuracy (~2 ULP) in the range of [0.09, 1], which is typically required for the conversion.

julia> Colors.pow12_5(0.5), Float64(0.5^big"2.4")
(0.18946457081379978, 0.18946457081379975)

julia> Colors.pow12_5(1.0), Float64(1.0^big"2.4")
(1.0, 1.0)

julia> Colors.pow12_5(1.5), Float64(1.5^big"2.4")
(2.6461778006805154, 2.6461778006805154)

julia> Colors.pow12_5(5.0), Float64(5.0^big"2.4")
(47.60126373669469, 47.591348467896964)

I decided to solve #425 first in order to do a fair benchmark for PR #482. However, it doesn't seem to be solved by just not using exp. 😕

If we can accept the divergence when the input is more than 2.0, then we can merge this. However, I would like to study #425 some more.

Edit:
@inline is not a good idea, but it could be a workaround for the problem with precompilation.

[codecov]

Codecov Report

Merging #483 (a601bc6) into master (c28d392) will increase coverage by 0.38%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #483      +/-   ##
==========================================
+ Coverage   92.41%   92.79%   +0.38%     
==========================================
  Files           9        9              
  Lines         975      985      +10     
==========================================
+ Hits          901      914      +13     
+ Misses         74       71       -3     

Impacted Files	Coverage Δ
src/parse.jl	100.00% <ø> (+2.52%)	⬆️
src/conversions.jl	99.33% <100.00%> (+<0.01%)	⬆️
src/utilities.jl	98.18% <100.00%> (+0.16%)	⬆️

---

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 6f0a3aa...a601bc6. Read the comment docs.

[kimikage]

Benchmark

julia> versioninfo()
Julia Version 1.6.1
Commit 6aaedecc44 (2021-04-23 05:59 UTC)
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i7-8565U CPU @ 1.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-11.0.1 (ORCJIT, skylake)

julia> xs = rand(1000) .+ 0.09;

julia> @noinline f(x) = Colors.pow12_5(x);

julia> @btime f.($xs);
  17.200 μs (1 allocation: 7.94 KiB) # before
  8.500 μs (1 allocation: 7.94 KiB)  # after