Remove cat adjoint in favour of ChainRules

[ToucheSir]

Now that this has been effectively superseded by https://github.com/JuliaDiff/ChainRules.jl/blob/v1.28.2/src/rulesets/Base/array.jl#L345, we can ease some of our maintenance burden. I've kept the tests around for now.

[mcabbott]

Sounds fine. Why not hcat, vcat, hvcat too?

[mzgubic]

Why not hcat, vcat, hvcat too?

IIRC there are some issues because Zygote adjoints are typed as Number rather than Real/Complex. But removing these out would be great!

[mcabbott]

I see rrule(::typeof(hcat), Xs::Union{AbstractArray, Number}...) there, vs. here:

@adjoint hcat(xs::Number...)
@adjoint hcat(A::Union{AbstractVector, AbstractMatrix}...)

In #1201 it was suggested to allow non-Number types too, which should be fine.

[DhairyaLGandhi]

Is there a plan on adressing the inferrability/ stability in the general case. With say multiple dimensions?