DEP: np.dtype('float64') == None and np.dtype(None) should be deprecated

[paris0120]

output: True
version: '1.18.5'
System: Windows 10

[eric-wieser]

Confirmed in 1.19.4

>>> np.dtype('float64') == None
True

thankfully

>>> np.dtype('float64') is None
False

I think this is unfortunately by design, since np.dtype(None) gives np.float64. Perhaps we should deprecate passing None, which would also make this comparison give a warning.

[seberg]

This ties in with other things about hashing and richcompare (e.g. gh-5345 and gh-).

But I think None specifically, together with np.dtype(None) itself could be deprecated. We have two versions for dtype conversions right now, one with default (accepting None) and one without. Apparently, these are two examples where we use the wrong version.

If anyone wants to look into this, i don't think it should be particularly difficult, although its all in C and requires adding deprecation tests, etc. so it is definitely not very easy either.