Adding an equivalent of in consistent with isequal

[nalimilan]

For CategoricalArrays's recode function, we bumped into the need for an equivalent of in which would always be consistent with isequal (rather than sometimes following == as it does for AbstractArray notably). This is in order to allow things like recode(array, [1, missing, NaN] => 0) (equivalent to replace(array, 1 => 0, missing => 0, NaN => 0)).

Currently, this works by special-casing an AbstractArray LHS to use isequal, but it was requested that this should work with other collections, such as intervals (JuliaData/CategoricalArrays.jl#290). Both Intervals.jl and IntervalSets.jl define in to be consistent with == (like AbstractArray), i.e. missing in 0..1 returns missing, so we cannot just use in. Yet we don't want packages to depend on each other.

It looks like we need a new function/method to complement in that would consistently use isequal (maybe waiting for a deeper change in Julia 2.0). How does adding such a function sound? I imagine that could be in(isequal, a, b), or a completely new function.

This issue was previously touched by discussion at #34744.

[bkamins]

Per the discussion on Slack, using isequal consistently for all in checks would make more sense in the future.

[milankl]

I find this

julia> NaN in [1, 2, NaN, 3]
false

counterintuitive and @oscardssmith questions whether really isequal should be used instead of == for in?