BUG: Avoid flattening single-colorant /DeviceN images to grayscale

[pablopupo]

Fixes #3302.

_get_image_mode has a /Separation branch that resolves its Pillow mode by recursing into the color space's alternate space, and a /DeviceN branch a few lines below that does mostly the same thing except for one special case. Any single colorant paired with a /DeviceCMYK alternate went straight to grayscale, warning "Color %s converted to Gray" for every colorant besides /Black. That shortcut was added in #2322 for K-only scanner images, where grayscale is exact, but it swallowed every other single-colorant DeviceN image the same way, which is what the reporter's /Cyan image was hitting.

The fix narrows that shortcut to require the colorant literally be /Black before returning grayscale directly. Everything else now falls through to the same recursive alternate-space lookup /Separation already uses (that recursion resolves a /DeviceCMYK alternate straight to CMYK on its own, no new codepath needed). I also checked whether _apply_decode's Separation-only decode-inversion override should be extended to DeviceN too, and it should not. _image_from_bytes's existing byte-replication fallback already gets the light-to-dark direction right on its own, and stacking Separation's decode reversal on top would double-invert it. Left it alone since it is a separate, pre-existing issue outside the scope of #3302.

I did not have a working copy of the reporter's PDF (an issuu.com viewer link with time-limited AWS download links, 50 MB), so I built a synthetic single-colorant /Cyan DeviceN CMYK image to verify. Before the fix, the extracted pixels for tints 0/128/255 come out (255,255,255), (127,127,127), (0,0,0), identical byte for byte to what a /Black-colorant image produces (mode L, plus the warning). After the fix, they come out (255,255,255), (63,63,63), (0,0,0) in mode CMYK with no warning. The /Black case is unchanged, still mode L, byte for byte.

Added two regression tests in tests/generic/test_image_xobject.py, one at the _get_image_mode level and one round-tripping through _xobj_to_image. Ran the full test_filters.py and tests/generic/ suites (162 passed, 5 pre-existing skips) plus ruff, mypy, and pyupgrade, all clean.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.88%. Comparing base (efc511b) to head (f5c371f).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3916      +/-   ##
==========================================
+ Coverage   97.86%   97.88%   +0.01%     
==========================================
  Files          57       57              
  Lines       10738    10736       -2     
  Branches     2006     2005       -1     
==========================================
  Hits        10509    10509              
+ Misses        127      126       -1     
+ Partials      102      101       -1     

☔ 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.