Add basic implementation for mypy: ignore comments

[ilevkivskyi]

Fixes #12358

Implementation is mostly straightforward, but there are few things:

• This requires Straighten type comment parsing mypyc/ast_serialize#95, so tests will fail until we merge that.
• We forgot to bump cache version to 6. This is trivial however, functions now have a flag indicating if they have a type comment.
• I found a pretty bad bug in native parser: parse errors cannot be type-ignored even if they are non-blocking. Again, it is a one line change, so not making a separate PR for this.

Finally, there is one thing that bothers me a bit: we have few places where we mention words type: ignore in errors and notes. However, it would take a significant effort to reliably change them to mypy: ignore when needed. Essentially we would need to record provenance for each error code, and thread it through to many places (including to cache metas, for some technical reasons). So we have three options:

1. Do nothing about it. mypy: ignore really means mypy-specific type ignore, so it is fine to use type: ignore in error notes, if we actually have mypy: ignore written in code.
2. Actually implement all the work needed (not a fan of this option TBH).
3. Change the wording in those few errors/notes so that it is more neutral: i.e. say Unused "ignore" comment instead of Unused "type: ignore" comment etc.

I am leaning towards option 3, but wanted to double-check before actually doing this.

cc @JukkaL

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

comtypes (https://github.com/enthought/comtypes)
+ comtypes/_post_coinit/misc.py:27: error: Unused "type: ignore" comment  [unused-ignore]
+ comtypes/__init__.py:26: error: Unused "type: ignore" comment  [unused-ignore]
+ comtypes/__init__.py:34: error: Unused "type: ignore" comment  [unused-ignore]
+ comtypes/connectionpoints.py:10: error: Unused "type: ignore" comment  [unused-ignore]

pydantic (https://github.com/pydantic/pydantic)
- pydantic/_internal/_utils.py:347: error: "Obj" has no attribute "copy"  [attr-defined]

rotki (https://github.com/rotki/rotki)
+ rotkehlchen/chain/ethereum/modules/eth2/eth2.py:681: error: Unused "type: ignore" comment  [unused-ignore]

pandas-stubs (https://github.com/pandas-dev/pandas-stubs)
+ tests/test_windowing.py:366: note: Error code "syntax" not covered by "type: ignore[arg-type]" comment

xarray (https://github.com/pydata/xarray)
+ xarray/tests/test_array_api.py:11: error: Unused "type: ignore" comment  [unused-ignore]

sockeye (https://github.com/awslabs/sockeye)
+ sockeye/layers.py:642: error: Unused "type: ignore" comment  [unused-ignore]

scipy-stubs (https://github.com/scipy/scipy-stubs)
- tests/misc/test_submodules_empty.pyi:3: note: See https://mypy.readthedocs.io/en/stable/running_mypy.html#missing-imports

[ilevkivskyi]

Primer looks good. All changes look like a change towards (most likely) desired behavior. In particular, there are few cases like this:

    some(code)  # mypy can't handle this  # type: ignore[arg-type]

while in fact mypy can now handle this, so the ignore is unused (and zero cases of the opposite, i.e. # avoid another # type: ignore).

[A5rocks]

Basically I just have a couple nitpicks.

[A5rocks]

I know the old code had the same thing, but I think this should have the original prefix, i.e. if I get an error for # type: ignore[mypy:fake] I expect to get Unrecognized error code "mypy:fake".

[A5rocks]

IMO we should be explicit about the prefix. I for one didn't know you could prefix error codes! i.e. Error codes in "mypy: ignore" should not be prefixed with "mypy:".

[A5rocks]

Sorry, I don't really know the parser and taking a look didn't really help. Does None mean # type: ignore? If so, I imagine the following blanket case should also apply here!