Do not block on duplicate base classes

[sobolevn]

@JelleZijlstra I don't really have a prefence on this issue. I am fine with both blocking and non-blocking errors here.

But, it looks like that making this non-blocking is quite straight-forward.
Closes #15349

[JelleZijlstra]

That's pretty unfortunate, can we make it inherit from Any or just skip the duplicate base instead?

[sobolevn]

Yeap, got it. Fixed!

[hauntsaninja]

This looks good to me!

One possible improvement here could be to deduplicate the base classes when calculating MRO, so instead of Any MRO we still get the MRO as if the base class was unduplicated.

[sobolevn]

@hauntsaninja I don't fill comfortable doing this. Because we can have two generic bases: Some[int] and Some[str] - which one we should remove? This way we can generate strange errors that won't help much. Any is the default way to handle errors inside, so let's stick to it.

[hauntsaninja]

Okay! I went ahead and added a test case for that situation

[github-actions]

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

vision (https://github.com/pytorch/vision) got 1.36x slower (40.8s -> 55.5s)