Remove bogus TODO added in #18585

[sterliakov]

Removes todo added in #18585 - such normalization is not technically correct when a tuple is used as a base class.

[ilevkivskyi]

Oh, sorry, I completely forgot about one important case where fallbacks can't be ignored: tuple subclasses. For example when one does:

class C(tuple[int, str]):
    ...

we must put tuple[int | str, ...] as an actual base Instance (because tuple is not variadic in typeshed, mostly for historical reasons, and this is unlikely to change).

[ilevkivskyi]

I think you can repurpose this PR to simply remove the TODO comment.

[sterliakov]

we must put tuple[int | str, ...] as an actual base Instance

How does/can that manifest and why aren't any testcases failing?

[ilevkivskyi]

First, all user-defined tuple types get patched in-place after semantic analysis (see calculate_tuple_fallback() and how it is called); second, tuple types are heavily special-cased in many common contexts. So it will probably only manifest in some obscure cases, but even then I don't want to add some code that is misleading.

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

freqtrade (https://github.com/freqtrade/freqtrade): 1.94x slower (131.3s -> 254.1s in a single noisy sample)