[Feature request] Add nanptp function

[Gabriel-p]

There is a ptp function but it will fail when there are nan values in the array.

[yliapis]

Hi Gabriel, I will take a go at this over the weekend

[rgommers]

Having this linked somewhere would be nice, but I think it's too niche to include in numpy. we only want nan* versions of the most commonly used functions, and ptp is not one of those.

[yliapis]

From looking at the existing nan* functions, it looks like most of the base functions have nan analogs. I agree that ptp is not the most commonly used function, but I use it quite a bit on signal processing problems, and it does feel like it is missing given the other functions in nanfunctions.py. I've finished implementing a basic version, I'll get it up on my account once it is cleaned up and tested.

[yliapis]

If we decide to move forward with this one, I just opened a PR with the function implementation and corresponding tests.
#13220

[rgommers]

We discussed this in the triage meeting. Most people were happy with rejecting this feature request. @charris abstained, @seberg said he could imagine a separate namespace.

The main rationale here is that we don't want more nan-functions in general; we have the most commonly used ones, and filling up the main namespace with more nan-functions isn't justified (there's a significant cost for adding anything to the namespace at this point.

In particular for this one: ptp is already niche, and was arguably a mistake to add to numpy. So nanptp isn't desired.

A separate package like bottleneck makes more sense for more nan-functions.

Thanks for suggesting this @Gabriel-p, and for the PR @yliapis

[Gabriel-p]

Thank you for the detailed explanation @rgommers