Mypy should detect obvious UnboundLocalErrors (e.g. declarations missing from an else: branch)

[lukasjuhrich]

Feature

If a variable is only declared in an if branch, but used afterwards, mypy should throw an error.

Pitch

Currently, the following snippet passes mypy --strict (v0.971/py3.9):

import typing as t
def foo(val: t.Optional[str] = None) -> int:
    if val is not None:
        derived_val = 1
    reveal_type(derived_val)
    return derived_val

Additionally, it reveals builtins.int for a clearly underspecified type.
This is pretty surprising, and in fact I did not catch an obvious bug because I mistakenly remembered that mypy had warned me about such issues in the past – but also because I assumed imypy could not possibly infer int in such a case: If I had a conflicting declaration in the else block, mypy would have already complained! So to some extent one might say this violates the principle of least surprise.

Relates to #686 and #6233 (specifically this comment).
(I hope this is not a duplicate)

[JelleZijlstra]

Duplicate of #686 and #4019.