Refactor `inverse_dict_type`

[goerz]

From @mofeing in #25 (comment):

From my side, one thing left that we should take care of before v1.0.0 is this method:

Bijections.jl/src/Bijections.jl

Lines 159 to 162 in 3522d66

	function inverse_dict_type(D::Type{<:AbstractDict{K,V}}) where {K,V}
	@warn "Using the default `inverse_dict_type` for $D. This may not be optimal for your specific dictionary type."
	D.name.wrapper{V,K}
	end

It does some internal hacking we shouldn't rely on. I added it because it was useful for BijectiveDicts.jl, but now that we are taking it more seriously, we should remove it and write extensions to that method for different dict types.

---

Might be good to put this into a v1.0.0 milestone to keep track of the things we still want to do before releasing v1.0.0 – it doesn't seem like like I have sufficient access to create milestones

[goerz]

I had a look at this, since I ran into the same issue in #41.

We could of course define extensions for some common packages that define AbstractDicts. I think that would those in DataStructures.jl and Dictionaries.jl.

But ultimately, we can't cover every implementation of AbstractDict that might exist in the ecosystem. You still need that fallback above. AbstractDict{V,K} (the fact that it takes exactly two type parameters) is part of the AbstractDict contract, so this should be fairly robust. I can't think of any way to get around the use of the internal D.name.wrapper. It should be fine… I'm not too worried about it going away. It's an internal that's been fairly widely publicized and there's JuliaLang/julia#35543 about the fundamental issue.

I'd be okay with closing this issue for v1.0.0 (keeping the existing function). And then maybe add extension for some external packages if the need arises, post-v1.0

[mofeing]

I see. Right now depends on Julia internals but in the case that it changes or a new feature that fixes this appears, it wouldn't break our API.

I agree with you and closing this for v1.0.0.

The only critical thing left for the big release would be #35 (#36 and #43 can be addressed before the release, but they are not holding it).