Automatically run MCprep tests

[TheDuckCow]

It will greatly help reliability in testing and PR review if we can have automated, remotely ran unit+integration tests. By doing so, we (ahem, me in particular) can be faster and more confident about approving code reviews, and we can avoid the situation where a review comes some days later only to say that there's a syntax error (not a good experience for anyone involved).

However, we will also need to be smart and manage our quota so that we don't burn through all 3K monthly minutes (too quickly, anyways). So, the proposal:

• Use the ubuntu low cost runner that github provides.
• Aim to just support the latest stable version of blender. Probably best if we hard code this so that we can manage when we're ready to test new versions. If we find we are well enough under quota limits, we could create additional params to also support a few other key versions of blender (e.g. 2.80 to ensure we don't drop downwards compatibility)
• To avoid having to download blender every single time (thus blowing through minutes faster), we should aim to use githubs artifact caching for the version of blender being targeted.
• Ideally we can post the test results back to the PR itself (something I've managed to do in other repos, very convenient for the author so they don't have to go hunting for the errors).
• Once functional and we can ensure working fine, this should be a new blocker for any PRs that touch python files.

[TheDuckCow]

In terms of how to execute, I was initially thinking we would just hard code the URL versions we want to use (by string concatenation via https://download.blender.org/release/), but worth noting there is also this pip tool here we could consider adding to our workflow as its own job. There is also this one which is focussed on pytest and blender (linking to the specific section about CI integration, which even includes some snippets on how to do caching).