Bring full-stack tests back online

[mattwthompson]

The last release of LAMMPS includes a test that lacks a decorator to gracefully skip when it's not installed. Might as well just add it to these tests, though.

[mattwthompson]

@ntBre does this mean anything to you? https://github.com/openforcefield/status/actions/runs/10011341975/job/27817227290#step:11:2300

These failures are

• only on M1 runners
• not obviously related to this changeset, but passing upstream: https://github.com/openforcefield/status/actions/runs/10051121052/job/27780177248#step:10:574

[ntBre]

Hmm, yes, that's coming from my tests for the new caching behavior. I added a context manager to temporarily disable the PortalClient's requests.Session field to ensure that the cached data was loaded from the local cache instead of being re-fetched. I see that the root cause is likely a sqlite3.OperationalError: attempt to write a readonly database, which leads me to the qcportal version since 0.55 replaced sqlite3 with apsw. We've been testing against 0.56 in the qcsubmit tests, and only on macos-12. The macos-12 images here are also pulling in 0.56, so I'm guessing something about the M1/0.54 combination is causing problems?

[mattwthompson]

I have two ideas here

1. Just downpin to some version(s) of things that work (0.54 across the board for QCArchive things, and some comptaible QCSubmit?). This pushes the problem down the road, which might be a value add if there is a genuine issue here
2. Add a lower constraint to keep tests running along. (Probably something like %s/~=0.50/~=0.56/ here). I prefer this if it works, but don't know that it will

[mattwthompson]

I don't know anything about QCPortal/QCSubmit compatibility in these version ranges, this is just the most recent release

[mattwthompson]

Okay, so qcportal >=0.55 doesn't work on M1 at all, but I'd like to avoid downpins. For now I think we should skip QCSubmit on M1 (only). This allows us to keep most M1 tests online without dropping QCSubmit on Intel hardware

[j-wags]

(to shortcut my search when I come back here in the future) Matt's opened an issue about the lack of availability of ARM builds for QCFractal here

[j-wags]

Thanks @mattwthompson - I agree that using the latest deps and removing M1 macs from the test matrix is the best option here.