[WIP] Make store return JSON-LD when preferred

[Treora]

The follow-up on #83.

Note that render does not return a string or response object, but an
object that jsonify can process. This can be changed when we wish to
support formats that are not JSON-based.

Also the content-negotiation/rendering is still only done on endpoints
that are read-only, because to use writable endpoints one needs to
understand the annotator-specific format anyway.

[BigBlueHat]

@nickstenning @tilgovi since @Treora's cycles are diminishing, do you think we could get this merged?

I'm happy to pick up where he left of if it needs any more tweaking.

Thanks for building this @Treora!

[Treora]

@BigBlueHat: I would like to note that this code never left the experimental phase. We're mostly talking about PR #83 here by the way, as this PR #90 is only adding content negotiation on top of that.
I'm not too fond of how it constructs json-ld directly using dicts, but I don't know if there is a satisfying rdf library to use instead. I've also assumed a namespace (http://annotatorjs.org/ns/) which is only defined implicitly, by its usage in this code.
But all in all, I suppose it provides a basic implementation of the (still applicable?) OA draft spec. So if the goal is just to have some working example, just rebasing and merging could just do the trick.