type datetimeindex constructor

[MarcoGorelli]

• Closes #xxxx (Replace xxxx with the Github issue number)
• Tests added: Please use assert_type() to assert the type of any return value

[Dr-Irv]

caught one issue regarding MultiIndex

[Dr-Irv]

We have an opportunity for cleanup here. No need to have both __new__() and __init__(). We should just have __init__(). We use __new__() instead of __init__() in the stubs when we want to return different subclasses based on the arguments. But for MultiIndex, that isn't the case. So we can delete __new__().

Ideally, we'd have types for all of the arguments, which will have to be based on the 3.0 docs, because the docs did get updated to correspond to what is in the code (which doesn't have _set_identity.

For this PR, I'm OK if you don't do them all, BUT names and name need to be SequenceNotStr[Hashable]

[MarcoGorelli]

thanks Irv! sure, I corrected name for multiindex, sorry for not having spotted that

[MarcoGorelli]

Looks like we might need MultiIndex.__new__ too, else the test I added fails pyright with

  /home/marcogorelli/pandas-stubs-dev/tests/test_indexes.py:70:62 - error: Argument of type "list[str]" cannot be assigned to parameter "name" of type "Hashable" in function "__new__"
    "list[str]" is incompatible with protocol "Hashable"

[Dr-Irv]

Interesting. I guess because the parent class uses __new__(), then the subclass has to do it as well. So for MultiIndex, we should provide __new__() and not provide __init__()

[Dr-Irv]

The only thing left to do is to remove __init__() from MultiIndex

[Dr-Irv]

thanks @MarcoGorelli