imgproc: fix calcBackProject() ignoring the 3rd and further channels - #29941

Fixes #29938

Problem

cv2.calcBackProject() returns the same value for pixels that differ only in
the 3rd channel, so a 3D (BGR) back-projection is wrong. From the issue:

roi  = np.array([[[10, 20, 30]]], dtype=np.uint8)
hist = cv2.calcHist([roi], [0,1,2], None, [8,8,8], [0,256,0,256,0,256])

img = np.array([[[10, 20, 30], [10, 20, 130]]], dtype=np.uint8)
bp  = cv2.calcBackProject([img], [0,1,2], hist, [0,256,0,256,0,256], 255.0)
# bp[0,0] = 255 (correct), bp[0,1] = 255 -- should be 0

Cause

Not in the back-projection kernel: calcBackProj_8u() handles dims == 3
correctly (tab[*p0] + tab[*p1 + 256] + tab[*p2 + 512]). That branch is simply
never reached.

A (8, 8, 8) ndarray is converted to an 8x8 CV_32FC8 Mat, not a
3-dimensional one (cv2_convert.cpp, ismultichannel = ndims == 3). The
std::vector<> overload of calcBackProject() already compensates: when
hcn > 1 it rebuilds the histogram as H, carrying the extra dimension. It
then validates against H -- which is why dims == 3 and the rsz == dims*2
assert pass -- but forwards the original hist to the implementation, which
re-reads dims from it, sees 2, and takes the 2D path. The third channel is
never read.

This also explains why the call does not simply throw: the checks run against
the corrected histogram while the computation runs against the uncorrected one.

Fix

Forward H rather than hist. H exists for exactly this purpose, and when
hcn == 1 it aliases hist, so single-channel histograms are bit-identical to
before.

Verification

Built imgproc + Python bindings, GCC 16.1 (MSYS2 UCRT64), Release, Windows.

Reproducer from the issue, before -> after:

opencv_test_imgproc, same binary before and after the one-line change:

unpatched  tests=2539  passed=2259  failed=257
patched    tests=2539  passed=2260  failed=256

Comparing the two failure sets, the only difference is the new test moving from
failed to passed; nothing else changed. (The 256 remaining failures are
pre-existing here for lack of a local opencv_extra checkout, and are identical
in both runs.)

The new C++ test was confirmed to fail without the fix and pass with it.

Tests added

• Imgproc_Hist_Calc.calcBackProject_3d_regression_29938 in
  modules/imgproc/test/test_histograms.cpp
• test_calcBackProject_3d in
  modules/imgproc/misc/python/test/test_imgproc.py (the issue's own case)

Notes

• Targeted at 4.x; 5.x carries byte-identical code and is affected the same way.
• modules/imgproc public API and ABI are unchanged.
• The OpenCL path is unaffected: CV_OCL_RUN is gated on hist.dims() <= 2 and
  returns before the changed line.

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

• I agree to contribute to the project under Apache 2 License.
• To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
• The PR is proposed to the proper branch
• There is a reference to the original bug report and related work
• There is accuracy test, performance test and test data in opencv_extra repository, if applicable
• The feature is well documented and sample code can be built with the project CMake