wcs.cdelt setter silently discards the unit of u.Quantity assignments #20500

Description

wcs's cdelt, crval and cd setters pass their argument through set_double_array, which converts any array-like to doubles. A Quantity therefore has its unit silently discarded and its magnitude reinterpreted in whatever cunit currently is.

This isn't new and isn't specific to the preserve_units kwarg, but preserve_units makes it much easier to hit, because it changes which unit is the "right" one. On the default path a WCS built from an arcsec header has been normalised to degrees, so cdelt = 1 * u.arcsec means one degree; with preserve_units=True the same WCS reports arcsec, so cdelt = 1 * u.deg means one arcsecond.

Expected behavior

I expected that, since the setter happily accepts a u.Quantity that it would also be using the unit in the calculation to set the cdelt. It should either reject quantities as input or convert them in a unit-aware fashion.

How to Reproduce

import astropy.units as u
from astropy.wcs import WCS

header = {"CTYPE1": "HPLN-TAN", "CTYPE2": "HPLT-TAN",
          "CUNIT1": "arcsec",   "CUNIT2": "arcsec",
          "CDELT1": 2.0, "CDELT2": 2.0,
          "CRPIX1": 1.0, "CRPIX2": 1.0, "CRVAL1": 0.0, "CRVAL2": 0.0}

# Default: astropy normalises the WCS to degrees
w = WCS(header)
w.wcs.cdelt = [1, 1] * u.arcsec
print(w.wcs.cdelt[0], w.wcs.cunit[0])     # 1.0 deg <- but asked for arcsec

# preserve_units=True: the WCS stays in arcsec
w = WCS(header, preserve_units=True)
w.wcs.cdelt = [1, 1] * u.deg
print(w.wcs.cdelt[0], w.wcs.cunit[0])     # 1.0 arcsec <- asked for degrees

Versions

platform
--------
platform.platform() = 'Linux-6.6.87.2-microsoft-standard-WSL2-x86_64-with-glibc2.35'
platform.version() = '#1 SMP PREEMPT_DYNAMIC Thu Jun  5 18:30:46 UTC 2025'
platform.python_version() = '3.13.8'

packages
--------
astropy              8.1.0.dev663+g1f930be7c.d20260928
numpy                2.5.3
scipy                1.17.1
matplotlib           3.10.9
pandas               3.0.5
pyerfa               2.0.1.5