Always try to report missing protocol members

[Gatsik]

Reporting missing members can't confuse user even more -- it can only help to see what's missing. Also, sometimes implementing missing member helps mypy to correctly infer the type it couldn't infer before

I'm not sure if a separate test is needed for that, but here's a silly example:

from typing import Protocol, reveal_type

class CMD[T](Protocol):
    def command(self) -> T:
        ...
    def arguments(self) -> list[T]:
        ...

class CMDImpl:
    def command(self) -> str:
        return ""
    # def arguments(self) -> list[str]:
    #     return []

class CMDRunner[T]:
    def __init__(self, cmd: CMD[T]) -> None:
        self.cmd = cmd


r = CMDRunner(CMDImpl())
reveal_type(r)

$ mypy t.py

t.py:20: error: Need type annotation for "r"  [var-annotated]
t.py:20: error: Argument 1 to "CMDRunner" has incompatible type "CMDImpl"; expected "CMD[Never]"  [arg-type]
t.py:21: note: Revealed type is "t.CMDRunner[Any]"
Found 2 errors in 1 file (checked 1 source file)

whereas after uncommenting missing protocol member implementation:

$ mypy t.py

t.py:21: note: Revealed type is "t.CMDRunner[builtins.str]"
Success: no issues found in 1 source file

and the note about this missing member would've had pointed to the cause:

t.py:20: error: Need type annotation for "r"  [var-annotated]
t.py:20: error: Argument 1 to "CMDRunner" has incompatible type "CMDImpl"; expected "CMD[Never]"  [arg-type]
t.py:20: note: "CMDImpl" is missing following "CMD" protocol member:
t.py:20: note:     arguments
t.py:21: note: Revealed type is "t.CMDRunner[Any]"
Found 2 errors in 1 file (checked 1 source file)

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅