Use error-hint for fft-related errors

[timholy]

Aqua was complaining about piracy. This eliminates some of the
piracy it was worried about, and the rest we temporarily (?)
circumvent by disabling Aqua's piracy test.

This package now only loads AbstractFFTs on Julia versions
that lack Base.Experimental.register_error_hint (Julia 1.4
and earlier).

[timholy]

I'm not wedded to this approach, so don't hold back with any concerns. We could also just disable the piracy test permanently.

If on the contrary you think this is an improvement, I can look into the other sources of piracy.

[codecov]

Codecov Report

Base: 62.67% // Head: 75.12% // Increases project coverage by +12.45% 🎉

Coverage data is based on head (9623cc1) compared to base (6fc57c1).
Patch coverage: 100.00% of modified lines in pull request are covered.

Additional details and impacted files

@@             Coverage Diff             @@
##           master     #190       +/-   ##
===========================================
+ Coverage   62.67%   75.12%   +12.45%     
===========================================
  Files          10       10               
  Lines         584      599       +15     
===========================================
+ Hits          366      450       +84     
+ Misses        218      149       -69     

Impacted Files	Coverage Δ
src/ImageCore.jl	78.37% <100.00%> (+5.96%)	⬆️
src/functions.jl	0.00% <0.00%> (-100.00%)	⬇️
src/stackedviews.jl	79.74% <0.00%> (+0.25%)	⬆️
src/convert_reinterpret.jl	64.44% <0.00%> (+4.44%)	⬆️
src/precompile.jl	87.77% <0.00%> (+87.77%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[johnnychen94]

My understanding is that type piracy will affect invalidation and thus hurt future using XXX time. Will this still be true for Julia 1.9?

I like this change -- registering error hints by type piracy works but isn't a good choice.
If anything, I think it's a better idea to improve the error hint mechanism -- make it easier to add hooked information for developers. Although I have no idea what could it be.

[timholy]

My understanding is that type piracy will affect invalidation and thus hurt future using XXX time. Will this still be true for Julia 1.9?

Yes and no. For MethodInstances that are well-inferred, the only way piracy invalidates is if you overwrite a method with the same type-signature (you're directly replacing a call in the dispatch chain). For MethodInstances that have poorly-inferred calls (e.g., fft(::AbstractMatrix), then any new overlapping definition is at risk for invalidating whether it's piracy or not. So yes, piracy is more of an invalidation threat than non-piracy, but the way in which it's more of a threat is fairly specific.

I don't know of any packages that define fft(::AbstractArray{<:Colorant}), so in this case invalidation of well-inferred code is probably not very likely.

[timholy]

In my view this is not a breaking change (it only changes the kind of error thrown), so does not require a breaking release.