Flux Optimizers should define equality

[oxinabox]

I was writing some tests, and I wanted to be sure that the hyper-parameters of two models were the same.
So I was checking equality on an object that contained all my hyperparameters,
including the optimizer (which has its own hyper-parameters).

Turns out that fails as

julia> Flux.Optimise.NADAM(0.003, (0.9, 0.9), IdDict{Any,Any}()) == Flux.Optimise.NADAM(0.003, (0.9, 0.9), IdDict{Any,Any}())
false

[DhairyaLGandhi]

Should be much easier as we remove the states completely from the optimisers.

[MikeInnes]

Yeah, this is not well defined right now but can happen as part of #637.

[ToucheSir]

#637 has since been spun off as Optimisers.jl, which supports equality comparison OOTB :)