don't autofix noqa'd errors, add --disable-noqa, add non-passing test for noqa reliability

[jakkdl]

big refactor for noqa functionality fixing most of the remaining ones in #185, but as I added the test for all errors being properly noqa'd and was greeted with a wall of errors I realized it might be good to commit and push before tackling them. If it's a minor fix I'll just push/amend it to this PR, otherwise open a separate one.

• Moves # noqa logic into Flake8TrioVisitor_cst, other than for ast visitors where some logic still is in __init__.py
  • utility visitor NoqaHandler removed
• add --disable-noqa, and use that instead of whether we're standalone to decide whether to parse noqa or not.
• the list of noqas is no longer parsed once by a utility visitor, but instead parsed by each visitor. This is a performance hit (though probably not very large, if any) but is sorta necessary as autofixing messes up line numbers. autofixing does not mess up line numbers in fact.
• should_autofix now takes a CSTNode parameter so it can determine the line number of it, and check if it's noqa'd
• visitor91x now doesn't autofix simple noqa scenarios, but doesn't handle loops and the like visitor91x handles all cases
• tests and shit

[jakkdl]

turns out fixing the errors made the PR simpler, as I had mistaken assumptions about how lineno-parsing was handled in CSTTransformers.

[jakkdl]

uh, is it an update to trio or pyright that has it throwing new errors? It now seems to be analyzing trio properly for some reason and seeing missing objects and whether functions are coroutines.

[Zac-HD]

I don't know why pyright has changed, but short-term I'd just ignore tests/ and longer term #187.

[jakkdl]

I don't know why pyright has changed, but short-term I'd just ignore tests/ and longer term #187.

added commit which excludes tests/eval_files and tests/autofix_files

[jakkdl]

I don't know why pyright has changed, but short-term I'd just ignore tests/ and longer term #187.

also related to #160