Move CategoricalArrays.jl into extension

[MilesCranmer]

Fixes #167. This should be completely backwards compatible. It just decreases load time if the user is not using CategoricalArrays.jl in their application.

Before:

julia> @time_imports using LossFunctions
      1.6 ms  DataAPI
      8.1 ms  Missings
      1.0 ms  Statistics
     62.9 ms  CategoricalArrays
    108.0 ms  LossFunctions

After:

julia> @time_imports using LossFunctions
      0.9 ms  Statistics
      0.4 ms  Requires
      4.0 ms  LossFunctions

[codecov-commenter]

Codecov Report

Merging #168 (b42e0ea) into master (cfcdc6a) will decrease coverage by 0.36%. The diff coverage is 33.33%.

:exclamation: Your organization is not using the GitHub App Integration. As a result you may experience degraded service beginning May 15th. Please install the Github App Integration for your organization. Read more.

@@            Coverage Diff             @@
##           master     #168      +/-   ##
==========================================
- Coverage   96.38%   96.03%   -0.36%     
==========================================
  Files           9       10       +1     
  Lines         526      529       +3     
==========================================
+ Hits          507      508       +1     
- Misses         19       21       +2     

Impacted Files	Coverage Δ
src/LossFunctions.jl	100.00% <ø> (ø)
src/losses.jl	100.00% <ø> (ø)
ext/LossFunctionsCategoricalArraysExt.jl	33.33% <33.33%> (ø)
src/losses/other.jl	91.11% <33.33%> (ø)

[MilesCranmer]

Tested on 1.9 and 1.6 – everything seems to work.

[juliohm]

Can we actually drop versions < v1.9? Do you have requirements to stick to older Julia versions? We are updating the entire stack to require Julia v1.9, it would be nice to do it here already.

[MilesCranmer]

I don't recommend that, as Julia 1.6.* is under long-term support for at least the next few years (LTS happens on a five year cycle): https://julialang.org/downloads/#long_term_support_release. Until a new LTS is released I would advise making key dependencies like this one backwards compatible. You can use Compat.jl for using new features on old Julia versions FYI.

A lot of packages depend on LossFunctions.jl and strive to have backwards compatibility to the LTS version. e.g., if the Julia version is fixed at 1.9 here I would probably need to drop the LossFunctions.jl dependency in SymbolicRegression because some of my users are still using 1.7 (for whatever reason; maybe their cluster uses it for stability?).

[juliohm]

It is ok to keep the PR as is 👍🏽 Many ecosystems are already requiring Julia v1.9, so I wondered if we could clean up the Requires.jl boilerplate here as well.