Rework handling of cancelled 'yield' in fixtures

[njsmith]

• Always report a failure if a fixture cancels the test, even if the
  fixture doesn't then raise an exception
• Inside fixtures, yield can now raise trio.Cancelled

Closes: #75, #77

[codecov]

Codecov Report

Merging #83 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master      #83      +/-   ##
==========================================
+ Coverage   99.78%   99.79%   +<.01%     
==========================================
  Files          19       19              
  Lines         476      498      +22     
  Branches       41       43       +2     
==========================================
+ Hits          475      497      +22     
  Misses          1        1

Impacted Files	Coverage Δ
pytest_trio/_tests/test_fixture_ordering.py	100% <100%> (ø)	⬆️
pytest_trio/_tests/test_fixture_mistakes.py	100% <100%> (ø)	⬆️
pytest_trio/plugin.py	99.5% <100%> (+0.02%)	⬆️

[njsmith]

@belm0 @touilleMan thanks for the 👍s! Now, who's going to hit merge? Our contributing docs say that people shouldn't merge their own PRs, so I guess it shouldn't be me...

[belm0]

@njsmith thanks for raising, I wasn't familiar with that policy.

In most projects I've been on, PR author merges after required approvals. It's because PR author knows the context best-- e.g. there may be a last minute correction or addition; the merge may need to be timed relative to others, etc.

[belm0]

... and ultimately person hitting merge button is responsible for build our runtime breakage. It seems like too much responsibility for a reviewer.

[belm0]

Another point I realized as I was going to merge this:

There is also decision about squashing, which really depends on whether commits were logically grouped or ad-hoc. It's best left to PR author. On my PR's I'd want that option.