Allow ... in some non-stub-file function bodies (protocol, abstract, etc)

[QEDady]

Describe the Bug

It seems that Pyrefly is permissive when using ... in functions or methods and totally reject pass e.g. in the following:

from typing import Protocol

class A(Protocol):
  def __foo__(self) -> int:
    pass
  def __bar__(self) -> int:
    ...

def foo() -> int:
    pass

def bar() -> int:
    ...

Pyrefly is ok with pass in both locations but rejects ... in both locations.

I think writing pass should be ok if used with protocols, abstract methods, .pyi, etc. On the other hand, I also think using ... should be an error outside of these abstract contexts. Other type checkers do that e.g. both ty and mypy would be ok with the protocol methods above but would flag the two functions after.

Sandbox Link

https://pyrefly.org/sandbox/?project=N4IgZglgNgpgziAXKOBDAdgEwEYHsAeAdAA4CeS4ATrgLYAEALqcROgOZ0Q3G6UN0AFag1wBjXFAA66aaKio4cOgEEAFENwjxUAJSJpdOphhg6AfTNhcuC6rgwoYHXQC0APk7oG%2B9IcPEFOAMjE3MzbFRKW3tHZ3dPb2DDQhTpaWNTK1xVOI9WRN8-AMU0rFCIyhzXPK8fPzoUwhAAGhAyShMoUkIRGigKAGJBUg6wLro0LDx8OnF0SDYAV0pUBghcdEJpIYBlGBg6AAsGBmI4RAB6C-bO7t42C5h0C8wxOAu5heXV9ee6K0odFQADdUNBUNhYLMNl8VmsNnRcMR4eg4Ft0GQGIcNi5gTBKHBfnQALx0SQgADMhAAjAAmcnSEAAX1aqFEazxADFoDAKJMcAQSOQmUA

(Only applicable for extension issues) IDE Information

No response

[randolf-scholz]

At least for stub files, using ... rather than pass is the established convention as per the typing spec: https://typing.python.org/en/latest/guides/writing_stubs.html#functions-and-methods

[yangdanny97]

We allow ... in stub files, but not in regular files. I think it makes sense to align with other checkers and allow ... in functions that can't be called.

I suspect the part that needs to be changed is the body_is_trivial definition. would be good to also add something that skips any docstrings at the top of the function body. is_ellipse does something like that

[yangdanny97]

I realized this is similar to #1916, we can fix them together