Fix device bug in subtract_dark(scale=True) misreported as unit mismatch

[mwcraig]

subtract_dark(scale=True) created the scale factor with a bare xp.asarray(scale_factor), which on a device-aware backend (array-api-strict on a non-default device, CuPy) lands on the default device and cannot be multiplied with the master dark. The resulting backend error was caught by the ValueError clause of the surrounding except and re-raised as a spurious UnitsError about mismatched units.

This PR creates the scale factor on the master dark's device (same pattern as gain_correct) and drops ValueError from the caught exceptions; the real unit check raises UnitsError from the CCDData wrapper, and test_subtract_dark_fails only relies on TypeError/UnitsError.

Test matrix (full pytest ccdproc):

backend	result
numpy	380 passed, 5 skipped
jax	366 passed, 10 skipped, 2 xfailed, 7 xpassed (pre-existing combiner/image_collection xpasses, unrelated)
dask	370 passed, 15 skipped
dask + CCDPROC_ENFORCE_ESCAPE_BASELINE=1	370 passed, 15 skipped

ccdproc/tests/test_ccdproc.py under array-api-strict: before 32 failed / 31 passed / 10 xfailed; after 28 failed / 35 passed / 10 xfailed. The four test_subtract_dark[...scale=True] variants go from failing to passing; no backend_xfail marker was present on them.

Fixes #966

🤖 Generated with Claude Code

https://claude.ai/code/session_01DTN9DnPnLKK2u7knnMJ2gA

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.83%. Comparing base (89e90f9) to head (d5cb7e1).
⚠️ Report is 7 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #973   +/-   ##
=======================================
  Coverage   95.83%   95.83%           
=======================================
  Files           8        8           
  Lines        1631     1631           
=======================================
  Hits         1563     1563           
  Misses         68       68           

☔ 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.