xfeatures2d: fix DAISY roi descriptor buffer overflow

[shimat]

ComputeDescriptorsInvoker computed the output row index as an absolute image offset (y*image->cols + x), but the descriptors buffer passed to DAISY::compute(image, roi, descriptors) is only roi.width*roi.height rows, addressed relative to the roi. Unless roi covers the whole image, this writes past the end of the allocated buffer.

Store the roi's y offset alongside the existing x offset and compute the row index relative to the roi. The absolute (y, x) passed to get_unnormalized_descriptor() is unchanged, since sampling is done against the full-image gradient layers regardless of roi.

Added a regression test in test_daisy.cpp comparing descriptors computed via a roi against the same pixels computed via the full image.

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
  Patch to opencv_extra has the same branch name.
• The feature is well documented and sample code can be built with the project CMake

[fallenmi]

AI disclosure: I used OpenAI Codex to assist this review; I verified the result locally against exact head 47e345fb70f4f515195423605ad158654ce2416f.

Looks good. The ROI-sized descriptor matrix now uses ROI-relative row indices while descriptor sampling remains in absolute image coordinates. I built the exact head against current OpenCV 5.x core and the new ROI/full-image equivalence regression passes.

As a deterministic control, I rolled back only the three production hunks while retaining the submitted test. The same test then fails across the ROI (maximum descriptor differences around 0.06 to 0.11); restoring the exact-head implementation makes it pass again. git diff --check is clean.

The retained upstream reds do not implicate this change: the Windows job exits before checkout because its support script is missing, while Ubuntu 26.04 fails unrelated GStreamer/G-API and BRISQUE tests. The other ten reported checks are green.