Unpacking a NamedTuple subclass uses superclass' types

If you subclass an existing NamedTuple to provide more specific type declarations for some of the fields, and then unpack an instance of that NamedTuple, mypy 0.770 gets its type information from the superclass, not the subclass. Minimal reproduction:

from typing import NamedTuple, Optional

class BaseInventory(NamedTuple):
    number: Optional[int]
    name: str


class Inventory(BaseInventory):
    number: int


item = Inventory(42, 'parrot')
item.number % 10  # mypy allows this
number, _ = item
number % 10  # mypy does not allow this

The last line reports:

error: Unsupported operand types for % ("None" and "int")  [operator]
note: Left operand is of type "Optional[int]"

I expected no error, since accessing the field by name works fine.

Start by running the minimal Python reproduction with mypy 0.770 and compare named-field access with tuple unpacking. Trace how NamedTuple subclass type information is used during unpacking; done when the unpacked number is inferred as int and the final modulo line no longer reports an error, with regression coverage for this reproduction.