Remove automatic `extern` in `isapprox`?

[nickrobinson251]

This may be the same issue as "do we want extern at all?" (JuliaDiff/ChainRulesCore.jl#56)

But I think if we have it, we may need it to be "smarter" if we're going to have isapprox(a::Differential, b) = isapprox(extern(a), b) as we currently do.

Because we will need to extern(a) (or maybe just unthunk?) the differential type before calling isapprox anyway if a it is not a scalar (e.g. the Zero Matix case in SVD tests: https://github.com/JuliaDiff/ChainRules.jl/blob/8c4985a9a941df165740badfbee83cdf93661742/test/rulesets/LinearAlgebra/factorization.jl#L20-L22)

[nickrobinson251]

we also call extern manually before isapprox in frule_test

[nickrobinson251]

#20 changes this to only have methods on AbstractThunk and AbstractZeros

[oxinabox]

we also call extern manually before isapprox in frule_test

I am inclined towards this and am infact doing it as part of solving #56 .
because isapprox on thunks gives error messages that don't include the numbers i need to see what is wrong.