fix: make test command histogram logic shape-agnostic for OpenCV 5

[faux123]

Bug

sudo howdy test crashes on systems with OpenCV 5.x:

  File "/usr/lib/howdy/cli.py", line 116, in <module>
    import cli.test
  File "/usr/lib/howdy/cli/test.py", line 131, in <module>
    hist_total = int(sum(hist)[0])
                     ~~~~~~~~~^^^
IndexError: invalid index to scalar variable.

Fixes #1127

Root cause

OpenCV 5.0 changed the Python binding of cv2.calcHist to return a 1-D array. With the 8-bin call in test.py it now returns shape (8,) where OpenCV 4.x returned (8, 1):

>>> import cv2, numpy
>>> cv2.__version__, numpy.__version__
('5.0.0', '2.5.1')
>>> f = numpy.random.randint(0, 255, (120, 160), dtype=numpy.uint8)
>>> cv2.calcHist([f], [0], None, [8], [0, 256]).shape
(8,)

With a 1-D array, built-in sum(hist) produces a numpy.float32 scalar instead of a length-1 array, so the [0] index raises IndexError. The subsequent value[0] in the bar-drawing loop fails the same way.

Fix

Flatten the histogram with ravel() and total it with hist.sum(), so the logic works with the array layout of every OpenCV version (4.x (8, 1), 5.x (8,)) instead of assuming either one. compare.py and add.py are unaffected: they already use np.sum(hist) and hist[0], which work with both shapes.

Environment where reproduced

• CachyOS (Arch-based), howdy master (d3ab993, via AUR howdy-git)
• opencv 5.0.0, numpy 2.5.1, Python 3.14.6, dlib 20.0.1
• Reporter in Howdy test results in IndexError: invalid index to scalar variable #1127 hit the identical traceback on Fedora with Python 3.14

Testing

• Patched logic exercised against real cv2.calcHist output under OpenCV 5.0.0, and synthetically against both the (8, 1) and (8,) layouts.
• Full howdy test loop (this exact patched file) run against a real IR camera for 30 frames on the system above: histogram bars computed, no errors, clean exit.
• Interactive sudo howdy test confirmed working with the equivalent fix applied to the installed copy.

[Raymo111]

@boltgolt can we please merge