Keep units in np.ptp with NumPy 2 - #281

With NumPy 2, np.ptp drops the units of a Quantity:

>>> import numpy as np, quantities as pq
>>> q = [1., 5., 3.] * pq.m
>>> np.ptp(q)
array(4.) * dimensionless     # expected array(4.) * m
>>> q.ptp()
array(4.) * m

On NumPy 1.26 this returned m, because np.ptp called a.ptp(). NumPy 2 removed ndarray.ptp, so np.ptp now runs maximum.reduce/minimum.reduce directly. Those reductions reach __array_wrap__ with context=None, so the result comes back dimensionless.

Fix: add a small __array_function__ that sends np.ptp to the existing Quantity.ptp when the first argument is a Quantity. Every other function still goes through ndarray.__array_function__. Quantity.ptp now also accepts keepdims, so np.ptp(q, axis=..., keepdims=True) works.

Tests: I added test_np_ptp to test_methods.py. It covers the whole array, axis given as a keyword and positionally, and keepdims.

• With NumPy 2.5 it fails without the fix and passes with it.
• The full suite passes with NumPy 2.5.4 and with NumPy 1.26.4: 190 passed, 3 xfailed.

I also added an entry to CHANGES.txt.

Separately, np.maximum.reduce(q), np.add.reduce(q) and np.add.accumulate(q) return dimensionless results on both NumPy 1 and 2. This PR doesn't change them. Fixing them would need ufunc-method-aware handling, for example in __array_ufunc__. I'm happy to look at that in a follow-up if it's wanted.

AI-assisted (Claude): I used it to help read the code and draft the test. I reviewed and ran everything myself.