Refactor GSD into a hard/soft dependency about garnett

[joaander]

@csadorf @bdice A future version of GSD will support reading in-memory buffers with the C implementation: glotzerlab/gsd#64

[csadorf]

How could a user use the python-only reader if desired?

[bdice]

@csadorf Can you define a use case where that would be needed? I don't know when the Python-only reader would be desired.

[csadorf]

The pure python module was introduced the allow reading of files that are not present on the file system, such as a binary (in memory) stream or a non-conventional file-like object such as a database stream.

In essence: the pure python module allows us to read files that have no filename.

[bdice]

@csadorf I guess that makes sense, though I am not aware of any use cases where that’s relevant except maybe a test/sample file in this package. Maybe there is a way to use GSD directly instead of copy-pasting the files into this package’s source.

[csadorf]

The complete polyhedra zoo is stored in a MongoDB database and is in theory accessible in this manner. Especially useful if you don't have access to /nfs/ as a collaborator for instance.

That said, you can get around it by writing to a temporary file on disk of course.

[bdice]

I'm changing my mind about this and I think the issue can be closed after #157. We now have a clearer picture of our soft/hard dependencies. Both the minimal and maximal set of dependencies are tested.