Allow method assignment in body of type extending typing.NamedTuple

this currently works for non-namedtuple classes (both runtime and mypy-time):

class C:
    x: int

    __hash__ = object.__hash__

and it works for typing.NamedTuple at runtime:

from typing import NamedTuple

class C(NamedTuple):
    x: int

    __hash__ = object.__hash__

$ python3 -i t2.py
>>> c = C(1)
>>> hash(c)
-9223363269847169460
>>> object.__hash__(c)
-9223363269847169460
>>> tuple.__hash__(c)
3430019387558

but fails at type-checking time:

$ mypy --version && python --version --version
mypy 0.770
Python 3.6.8 (default, Oct  7 2019, 12:59:55) 
[GCC 8.3.0]
$ mypy t2.py
t2.py:7: error: NamedTuple field name cannot start with an underscore: __hash__
t2.py:7: error: Invalid statement in NamedTuple definition; expected "field_name: field_type [= default]"
Found 2 errors in 1 file (checked 1 source file)

concrete usecase: __hash__ is in the critical path (~13%) for some code I'm writing and the extra layer of function which calls into object.__hash__'s overhead is significant:

# mypy is ok with this
from typing import NamedTuple

class C(NamedTuple):
    x: int

    def __hash__(self) -> int:
        return object.__hash__(self)

I'd be happy to hack on this given some pointers :)

Reproduce the issue with the t2.py examples and run mypy on the file to confirm the two diagnostics for hash. Trace how NamedTuple class bodies distinguish fields from method assignments, then make the assignment type-check while preserving validation for invalid field declarations.