Stripe identity off ComposedOptic when test equality

[sunxd3]

The method

function Base.:∘(vn::VarName{sym,<:Lens}, lens::Lens) where {sym}
    return VarName{sym}(getlens(vn) ∘ lens)
end

was removed with the #91.

The trigger was: Accessors changes the order of composition (Accessors follows function composition order, i.e., f ∘ g mean apply g first, which I think is more intuitive), so we needed to adjust the above function for the new order or remove it.
With my personal taste, I removed it (sorry for missing @torfjelde's comment here), because it strikes me a bit strange composing a VarName and an optic.

WIthout this function, In DynamicPPL, we will need to do something like VarName{getsym(vn)}(optic ∘ getoptic(vn)) explicitly, which works okay (modulo the extra code). (One current example in DynamicPPL using ∘ between VarName and Lens.)

But one issue surface:
Sometimes in DynamicPPL we need to retrieve a value from a dictionary like Dict(VarName{:a}(IndexLens(1)) => 1.0, ...), with varname VarName{:a}(IndexLens(1) ∘ identity). Just as Tor's concern, this will fail.

This PR implements a more allowing equality tests to circumvent this issue. But I am also happy to consider restoring the aforementioned Base.:∘(vn::VarName{sym,<:Lens}, lens::Lens) function.

[sunxd3]

@tor @devmotion @yebai opinions?

[yebai]

Probably better for @torfjelde to review this?

[sunxd3]

PR closed as this seems to be a dead end -- ComposedOptic has Inner and Outer type in the signature, so the Dict key type will never match even with == defined.