feat: relax opencv pin to support both 4.x and 5.x, add CI matrix axis

pyproject.toml's opencv-python/opencv-contrib-python pin was capped at
<5.0.0 since 2026-06-16, defensively, before the codebase actually
supported OpenCV 5. PR #61 landed real OpenCV 5 compatibility fixes
(aruco, HoughLinesP, BRISK/AKAZE, MSER) and release 2.2.0's changelog
already headlined "works under both OpenCV 4 and 5" -- but the pin
itself was never relaxed, so pip install could never actually resolve
opencv5, and CI never exercised it either.

Relax the pin to >=4.5,<6.0.0 per issue #44's plan, and add an
opencv-version axis (4, 5) to ci.yml's test matrix so both are
actually exercised on every push/PR, not just verified manually once.

Verified locally in isolated venvs (not the ambient dev environment):
full suite green against real opencv-contrib-python 4.14.0.94 (785
passed, 94 skipped) and 5.0.0.93 (784 passed, 95 skipped).

Fixes #44

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Author: petercorke

.github/workflows/ci.yml

@@ -15,6 +15,7 @@ jobs:
       matrix:
         os: [ubuntu-latest, macos-15, windows-latest]
         python-version: ["3.10", "3.11", "3.12", "3.13"]
+        opencv-version: ["4", "5"]
 
     steps:
       - uses: actions/checkout@v7
@@ -30,6 +31,7 @@ jobs:
       - name: Install package
         run: |
           pip install .[dev,tool]
+          pip install "opencv-python==${{ matrix.opencv-version }}.*" "opencv-contrib-python==${{ matrix.opencv-version }}.*"
           cd packages/mvtb-data && pip install .
 
       - name: Run tests

pyproject.toml

@@ -39,8 +39,8 @@ dependencies = [
     "scipy",
     "matplotlib",
     "pillow",
-    "opencv-python<5.0.0",
-    "opencv-contrib-python<5.0.0",
+    "opencv-python>=4.5,<6.0.0",
+    "opencv-contrib-python>=4.5,<6.0.0",
     "spatialmath-python",
     "pgraph-python",
     "ansitable",