Support for interleaved errors

[ikonst]

Here's what I came up with while trying to replicate mypy's own testcases, which interleave the expected errors in the source code like so:
https://github.com/ikonst/pynamodb-mypy/blob/9453404f576c4ead04749d61eb041e10bec13229/tests/test_plugin.py

The implementation is here:
https://github.com/ikonst/pynamodb-mypy/blob/9453404f576c4ead04749d61eb041e10bec13229/tests/mypy_helpers.py

In a failure case, the failed assertion output will look like:

>       assert actual == expected
E       assert 'from pynamod...ibute "lower"' == 'from pynamodb..._attr.lower()'
E           from pynamodb.attributes import UnicodeAttribute
E           from pynamodb.models import Model
E
E           class MyModel(Model):
E               my_attr = UnicodeAttribute(null=True)
E
E           reveal_type(MyModel.my_attr)  # E: Revealed type is 'pynamodb.attributes._NullableAttribute[pynamodb.attributes.UnicodeAttribute, builtins.str]'
E           reveal_type(MyModel().my_attr)  # E: Revealed type is 'Union[builtins.str*, None]'
E         - MyModel().my_attr.lower()  # E: Item "None" of "Optional[str]" has no attribute "lower"
E         + MyModel().my_attr.lower()

Of course, this requires assertion rewriting, which would've been done in the pytest plugin normally:
https://github.com/ikonst/pynamodb-mypy/blob/9453404f576c4ead04749d61eb041e10bec13229/tests/__init__.py#L3

There are two advantages of interleaving the errors in the test code:

1. changing the test code does not necessitate updating of line numbers
2. updating the test code is easier (by diffing the program against the "expected program", i.e. data-driven testing)

Have you considered allowing a similar approach?

[dmtucker]

I had not, but it's a neat idea!
I started playing with it locally a little... Seems like it could be an interesting/useful option 😊