fix: byteswap non-native input instead of reinterpreting it

[matthiasschabel]

Data in non-native byte order maps to the wrong colors.

• On a little-endian host, a big-endian float array has its bytes relabelled rather than
  reordered, so -0.5, 1.5 and NaN all decode as tiny denormals.
• They land on the first ramp color, which makes under, over and bad unreachable for
  such input.

import numpy as np
from cmap import Colormap, Color

cmap = Colormap(["red", "blue"], under="green", over="yellow", bad="black")
native = np.array([-0.5, 0.5, 1.5, np.nan])
non_native = native.astype(native.dtype.newbyteorder())

print([Color(c).hex for c in cmap(native)])
print([Color(c).hex for c in cmap(non_native)])

['#008000', '#7F0080', '#FFFF00', '#000000']   # under, ramp, over, bad
['#FF0000', '#FF0000', '#FF0000', '#FF0000']   # all the first ramp color

The numpy 2 migration in #60 replaced xa.byteswap().newbyteorder(), which numpy 2 removed,
with a plain .view() of the swapped dtype. byteswap().view(dtype.newbyteorder()) is the
numpy 2 spelling of the original, and what matplotlib uses today.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.70%. Comparing base (02fa073) to head (ff1d461).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #152      +/-   ##
==========================================
- Coverage   95.71%   95.70%   -0.01%     
==========================================
  Files         168      168              
  Lines        2192     2191       -1     
==========================================
- Hits         2098     2097       -1     
  Misses         94       94              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[tlambert03]

thanks!