Restructure pytest plugin hooks

[altendky]

These changes enable handling of nested async fixtures and non-function scope async fixtures.

WIP for:

• More self review for trivial cleanup
• Self review of architecture
• Matching fixture teardown order
  • Presently this PR is coded for concurrent teardown which sounds nice with async but some fixtures will not only have startup dependencies on other fixtures but also teardown dependencies. This should be able to be documented etc but for now we should probably just skip concurrent teardown. That said, there's a test for it so... hmm.
  • In another PR add config for this and deprecate the default concurrent teardown. Or maybe leave it and allow fixture teardown dependency specification. Or default non-concurrent and ... do it somewhere else. Proper ordering of async fixture setup and teardown including inter-fixture dependencies #57
• Attempt to get 'external' feedback
• Review and update readme
• Fix outcome assertion Consider all outcomes in assert_outcomes() #98
• Address loss of concurrent fixture teardown caused by 50983ee and masked by the bad outcome assertions

[altendky]

@cdunklau, I haven't made it all the way through my basic self-review let alone a reconsideration of where these changes evolved too... but I think this is may be worth looking at at this point and trying out. I'm thinking that it might be worth testing non-function scope fixtures as well. Should be easy enough with a fixture that yields itertools.count() or somesuch but I'm not quite going to get to that tonight. I just think it is worth looking at this as a restructure that implements a few missing pieces rather than just fixing your original issue.

[meejah]

Maybe I'm not following the code exactly, but the maybeDeferred in in_reactor later on should handle that, I think? (referring to the "TODO" comment)

[altendky]

And this is why I need these todo's so I come back and get a clue... :| I was thinking there were inlineCallbacks fixtures and they ought to be handled here too. There aren't. *smh* and such. Thanks.

Though, seems like they could be implemented (after 30 seconds of thought anyways). Add a decorator to mark them and intercept them here for setup before pytest gets confused by them.

[altendky]

Hi @vtitor and @schmir. After doing this PR I am starting to feel like I actually have some grasp on how pytest-twisted works overall but it is only reasonable to assume there is still plenty of history and reasons that I still don't know (or straight up errors I've made as well...). If you have time I would appreciate any feedback on these changes, even if just some quick questions or such about something that looks fishy. If not, no worries. I get it being busy with other things. :] As always, thanks for the work you already donated to this project.

[schmir]

Sorry, but I don't have time to look into this.

[altendky]

To have it in the record here: pytest-dev/pytest#3261 (comment)