Add negative and positive element-wise specs

[kgryte]

This PR

• adds specifications for the element-wise APIs: negative and positive.
• follows the conventions set forth by other element-wise APIs.

Notes

• While negative is relatively common across array libraries, positive is less common. The main impetus for this PR is to ensure functional API equivalents to the magic methods __neg__ and __pos__, respectively, as discussed, and endorsed, in consortium meetings.
• Torch uses the name neg, while NumPy et al and TensorFlow use negative. I've followed NumPy precedent here, rather than magic method conventions.

[rgommers]

Do we really need positive? I can't remember an explicit discussion on that function. Given that it's basically a do-nothing call, I'm not sure why anyone would want to use it. NumPy does have it I see, never noticed it before. bool input raises an error there, which makes sense - but that isn't specified in this PR.

[kgryte]

I suppose it depends if we believe that the function should be included for completeness. This is true both for __pos__ equivalence (if __pos__ should be defined to be with) and because negative is defined.

Thanks for pointing out the NumPy exception (and deprecation for __pos__ and non-numerical arrays). At minimum, I'll update this PR to indicate that both negative and positive should only support numerical arrays and not bool.

[rgommers]

I suppose it depends if we believe that the function should be included for completeness.

Yeah it does make sense from that perspective. If it's the one dunder method that would otherwise be missing, let's include it.

[rgommers]

Discussed in today's call, everyone seemed happy to also include positive. Merging.