DSLParser should not store function as state

[AlexWaygood]

In most methods of the DSLParser, self.function can be safely assumed to be an instance of Function. However, it cannot be assumed to be an instance of Function in all methods, because it is set to None in self.reset() before starting to parse a block:

cpython/Tools/clinic/clinic.py

Lines 4457 to 4458 in 49f238e

	def reset(self) -> None:
	self.function = None

This is a pretty fragile design, and means we have to have a bunch of assertions scattered about the DSLParser class in order to keep mypy satisfied:

cpython/Tools/clinic/clinic.py

Line 4879 in 49f238e

assert self.function is not None

cpython/Tools/clinic/clinic.py

Line 4884 in 49f238e

assert isinstance(self.function, Function)

cpython/Tools/clinic/clinic.py

Line 4922 in 49f238e

assert self.function is not None

cpython/Tools/clinic/clinic.py

Line 5292 in 49f238e

assert self.function and self.function.parameters

cpython/Tools/clinic/clinic.py

Line 5306 in 49f238e

assert self.function is not None

This is a pretty tell-tale sign that function shouldn't really be stored as state at all. Instead, it should be passed into these methods as parameters, and returned from these methods after having been modified. Unfortunately, I think getting there would be a pretty significant refactor...