Pin ujson<2; nbconvert<6

[dwhswenson]

• Pinning ujson < 2 resolves Error when loading a saved committor simulation storage file  #922. (@hejung -- this is correct, right?) Long term solution is that the new storage I'm working on doesn't require ujson.
• Pinning nbconvert < 6 fixes a problem in current builds of docs. Something fails and then hangs in our custom notebook conversion sphinx plugin (written by @jhprinz and no longer maintained). Long term solution is to move to nbsphinx.
• Removed pin on pytest 5.3.4. Still shouldn't use that specific version, but pytest is up to 6.0 now, so I see no need to keep a pin to avoid a single specific regression.

[hejung]

Yes, ujson < 2 works.
Anything to consider or anything specific I should look out for when I finally try out the new storage?

[dwhswenson]

Thanks @hejung: I'll consider that enough of a review to go ahead and merge this. 😄

Anything to consider or anything specific I should look out for when I finally try out the new storage?

New storage files will be completely incompatible with old storage files (but much faster). I'm hoping to put up some PRs very soon if you want to play with the dev version -- so far it's only for toy systems (and one branch has Gromacs support), but I'll add OpenMM support soon. @sroet has been giving it a stress test on toy systems for some method development he's doing. It will start out under openpathsampling.experimental.storage, so it can co-exist with the current storage.

From a developer perspective, the main difference comes if you implement a new engine (more correctly, if you implement new "snapshot features"). It also includes a complete rewrite of collective variables, so custom subclasses of CollectiveVariable would require some changes. All the cv_requires_lists, cv_time_reversible etc. are treated very differently (there are still shortcuts for those, but treated more like the "hooks" structure). Pretty much everything else should be the same: I'm using the same to_dict/from_dict API as before.

Plan is to get new storage in as an experimental thing, give it some stress tests, then it will become the main storage (and one of the major features) in 2.0.