This commit sets up the configuration for towncrier. We'll put changelog
fragments in the directory changelog.d, which is being created (or
technically a stub file to ensure it always exists), and they will be
accumulated into CHANGELOG.rst, which I think is a pretty common name
and is consistent with other changelog-related terminology. I've renamed
NEWS.rst to CHANGELOG.rst accordingly.

This hook just does some basic sanity checks on the changelog fragments.
I've chosen not to use the update hook because I'm not sure how it would
interact with setuptools-scm (which changes the version number on every
commit), and it might not even be a good fit for our workflow anyway.

I believe that jaraco.develop.finalize is not the right tool for us to
use when preparing for a release. For one thing, when I run it I get
errors about a missing dependency (tomlkit); probably more importantly,
it seems to do a few things that are incompatible with the way we have
this project set up, like creating an unused newsfragments directory and
adding some random files to Git. I haven't fully investigated this but I
think there is minimal benefit in using jaraco.develop.finalize compared
to just running towncrier directly.

So I've modified the tox environemnt to do just that. I changed it to
actually install the package to avoid and problems with missing
dependencies (although it's possible that wasn't the right way to fix it
but we can figure that out later) and I replaced the previous command
with a simple invocation of `towncrier build`. I also changed
the environment name to `towncrier` to reflect its new role.

Note that when preparing a release, we'll have to pass the --version
option to towncrier (or, to `tox run -e towncrier`) manually for now. I
hope that in the future we can automate that process.

This is supposed to be an example of how we should be handling changelog
fragments, but it also serves to document the change of enabling
towncrier. There should be one of these files in every pull request
going forward.