Update to devices.devcore to use a QubitGraph rather than an availability dict

[tjproct]

This circumvents a bug, currently in develop, whereby availability dicts aren't propagated to a QubitProcessorSpecs's QubitGraph. Although this won't need to change if that bug is fixed, this is probably a more robust way to do things.

[coreyostrove]

This looks good to me at first glance, but could you try merging develop into this branch? I took a skim through the unit testing logs and I think all of these errors are things that have been fixed in develop (most of them are due to calls to a deprecated numpy function which Erik fixed a couple months ago), but would want to make certain of that.

[tjproct]

Ah yes, hadn't pulled develop in a long time. Latest develop merged in. Thanks Corey!

[coreyostrove]

Thanks for merging that in. It looks like almost all of those errors were indeed fixed already on develop. The only remaining one I can see is TestDirectRBDesign on Windows with python 3.6 (3.7 works and I can't re-run the tests on 3.8 or 3.9 but they presumably work too). I'm guessing you may have written this unit test, so normally I'd ask that you look into it, but I've run into these sorts of OS/python version combination incompatibilities before and they are usually for obscure and tricky to track down reasons. So I don't mind looking into it first and pinging you if I can't figure it out.

That said, given the discussion in #265 do you want to table this PR temporarily in order to incorporate any related refactors?