Fix deployment tests, skip tests requiring optional dependencies

[mattwthompson]

This PR fixes a small typo in the deployment tests (that would have been quite difficult to see ahead of time in #803). This is contained to just this workflow; the release is not broken. When comparing git tags, the runner steps up one directory before running git ls-remote to find the tag of the cloned repo, which is then compared to the version found in the Python package. This check works fine and passes. But when going back into the repo, it needs to be openff-toolkit, not openff/toolkit. So this line is an exception to the general pattern of which path needs to be stepped into.

I'll see if I can run the workflow from this branch to be sure it works.

• Tag issue being addressed
• Add tests
• Update docstrings/documentation, if applicable
• Lint codebase
• Update changelog

[mattwthompson]

This link is probably not persistent, but the failing check is fixed and the tests are passing so far: https://github.com/openforcefield/openff-toolkit/runs/1729531265?check_suite_focus=true

[codecov]

Codecov Report

Merging #818 (d99459e) into master (6e11ed6) will decrease coverage by 0.14%.
The diff coverage is n/a.

[mattwthompson]

To clarify, this PR does two things

1. Fixes the openff-toolkit path while the runner is moving around in the directories
2. Marks some tests as xfail

# 1 will work before a release, but # 2 will not work until the next release because this action checks out the most recent commit. But at least the tests will run, and only fail for a small set of known "failures."

[j-wags]

Looks good, and the test_molecule changes are really nice too. For the tests where OE and RDKit gave insignificantly different results (eg different precisions), I've changed them to be separate tests. Thanks, @mattwthompson!