WIP: better citation processing

[John MacFarlane <jgm-TVLZxgkOlNX2fBVCVOL8/A@public.gmane.org>]

I've been working this summer on a replacement to pandoc-citeproc.

Part of this is a new library, citeproc, which is
not specific to pandoc. This already passes a larger portion
of the CSL test suite than pandoc-citeproc, and it should be
straightforward to improve it further. It will be easier to
maintain than pandoc-citeproc, more accurate, and faster.

I've used this library to create a new filter,
new-pandoc-citeproc, which now passes most of the pandoc-citeproc
test suite but runs around 6 times faster.

I'm inclined, though, not to release this as a new filter,
but instead to depend on the citeproc library and build the
citation processing capabilities into pandoc itself. This will
cut down the binaries we need to distribute from two to one, and
it will simplify things for users, who won't have to worry about
filters. It will also be more performant, as we'll avoid the
overhead of JSON serialization and deserialization.

I'm not quite ready to release any of this code, but I hope
to do so in the next month or two.  This is just a teaser.

The new library is pure Haskell and won't depend on bibutils
(a wrapper around a C library).  That means we'll only support
bibtex/biblatex, pandoc yaml, and CSL JSON as bibliography
formats.  Those who need others will have to convert them using
standalone bibutils.  But support for other formats was never
great, so I don't think this is a big loss.