Possible regression: a niche case in expression type detection

[neutrinoceros]

Bug Report

After upgrading mypy to 0.930 I'm finding new confusing errors that I think are incorrect (false positives)

To Reproduce

This example may not be strictly minimal, but it's close to the real case example where I discovered the new error, while still being much more concise.

# t.py
from typing import Tuple, List

ElementT = Tuple[str, Tuple[str, List[str]]]
ContainerT = Tuple[ElementT, ...]

a: ContainerT = tuple((f1, (f2, [])) for f1, f2 in (("", ""), ("", "")))

b: ContainerT = (("", ("", [])),)

# The following four lines should be equally valid (or invalid) regarding types
# but only the last one raises an error
c1: ContainerT = a + b  # works fine !
c2: ContainerT = b + a  # this too !
c3: ContainerT = a + (("", ("", [])),)  # also works
c4: ContainerT = (("", ("", [])),) + a  # but this doesn't

$ mypy t.py
t_typ.py:16: error: Incompatible types in assignment (expression has type "Tuple[Union[Tuple[str, Tuple[str, List[<nothing>]]], Tuple[str, Tuple[str, List[str]]]], ...]", variable has type "Tuple[Tuple[str, Tuple[str, List[str]]], ...]")
Found 1 error in 1 file (checked 1 source file)

Expected Behavior

No error (this example type-checks with mypy 0.910)

Your Environment

• Mypy version used: 0.910 and 0.930
• Mypy command-line flags: none
• Mypy configuration options from mypy.ini (and other config files): none
• Python version used: 3.10
• Operating system and version: MacOS 12.0.1