I have the following concern regarding the way I have changed the type annotation:
Sequences only consisting of None elements are now valid according to the type hint, but correctly lead to the error TypeError: no arrays or types available to calculate result type. I am not sure, however, whether/how it is possible to type-hint that at least one element needs to be of type ArrayLike. Since passing an empty sequence is also valid according to the type hint but also leads to an error (ValueError: Must provide at least one operand), maybe this is a similar case...

I leave it up to the maintainers to decide if it is more important to not raise a type-checking error on intended usage or to raise an error when passing a Sequence[None]-type value as op to nditer.

Sequences only consisting of None elements are now valid according to the type hint

I'm afraid that's inevitable, as there's no way to express "a Sequence[ArrayLike | None] that has at least 1 element that isn't of type None".

I added such a type test in cf3dc6b based on the information I found in the testing readme of the archived numpy-stubs repository as I could not find any information in the numpy repository itself. If this information is still up to date, maybe it should be migrated to this repository as well?

I added such a type test in cf3dc6b based on the information I found in the testing readme of the archived numpy-stubs repository as I could not find any information in the numpy repository itself. If this information is still up to date, maybe it should be migrated to this repository as well?

Yea that's a good point. There's a good chance we'll be overhauling the type-tests soon, so I suppose we'll take that into consideration once we do that :)

Thanks, Simon!