Drop nox in favor of vanilla scripts

[florimondmanca]

Fixes #239. Also based on some discussion @tomchristie and I had offline.

[florimondmanca]

Hmm… pytest is showing a "Could not load plugin" error:

pluggy.manager.PluginValidationError: Plugin 'trio' could not be loaded: (attrs 18.2.0 (/home/travis/virtualenv/python3.7.1/lib/python3.7/site-packages), Requirement.parse('attrs>=19.2.0'), {'trio'})!

This happened to me in asgi-lifespan earlier today. Probably due to a silly dependency resolution issue… 😕 Not sure why it's only happening now; there hasn't been a release in trio, attrs, pytest or pluggy recently that seems to change the behavior.

Anyway, forcing attrs>=19.2 should fix it on our end, but it's definitely a hack…

[florimondmanca]

It seems the Windows build is stalling, and has been since the addition of Unix socket support (#511 - see build here).

We didn't notice because the job was running last, and was allowed to fail, so the build was green even though the Windows job hadn't finished yet (and would finish minutes later).

My most obvious guess would be that Unix domain sockets… aren't available on Windows, but for some reason open_uds_stream() is stalling instead of failing.

I pushed a commit here to revert to the old situation, but we'd need to skip tests that use Unix sockets on Windows.

[tomchristie]

Looks great!
One minor question about how we ought to handle serving the docs vs building the docs.

[florimondmanca]

@tomchristie Are we good running with scripts/docs-{build,serve}?

[tomchristie]

@florimondmanca - Personally I'd probably prefer scripts/docs build and scripts/docs serve, but either way let's pull this in for now and we can figure out if there's anything else we want to iterate on later. 👍