Add unit tests for traincascade library

[vladiant]

Summary

Introduces a doctest-based unit test suite for the traincascade library, wires it into CI across Ubuntu / macOS / Windows, and registers the test target with the top-level CTest runner.

Motivation

The traincascade library previously had a doctest harness scaffolded (main.cpp) but no actual test cases. This MR fills that gap with a comprehensive unit test suite focused on the library's pure / configuration surface, and ensures the tests run automatically on every push.

Changes

New test files

File	Coverage area
test_o_utils.cpp	icvCmpIntegers, cvAlign, CV_DTREE_CAT_DIR, LessThanIdx, LessThanPtr, cvPreprocessIndexArray, BlockedRange / parallel_for / parallel_reduce, CvDTreeNode::get_num_valid / set_num_valid
test_params.cpp	CvDTreeParams, CvBoostParams, CvCascadeBoostParams, CvCascadeParams (default + parameterized ctors, scanAttr)
test_features.cpp	CvFeatureParams::create, CvFeatureEvaluator::create, CvHaarFeatureParams / CvLBPFeatureParams / CvHOGFeatureParams, evaluator init / generateFeatures / setImage
test_imagestorage.cpp	CvCascadeImageReader::create failure paths

All tests follow the Arrange–Act–Assert pattern with explicit comments and cover both the happy path and edge cases (out-of-range indices, duplicates, multi-dimensional inputs, unsupported types, empty ranges, null vs. allocated num_valid, too-small HOG window, unknown attribute names, missing files, …).

Build system

• CMakeLists.txt
  • Registered the four new test sources on the test_traincascade target.
  • Removed -Wold-style-cast, -Wsign-conversion, -Wconversion from the test target only — these are triggered by inline definitions in the library's public headers (haarfeatures.h, lbpfeatures.h, HOGfeatures.h, traincascade_features.h) which the library itself does not enable.
• CMakeLists.txt
  • Added include(CTest) + enable_testing() at the project root so ctest discovers the test from the top-level build directory.

CI

Added a Test step after the build step in:

• ubuntu.yml
• macos.yml
• windows.yml

Each runs ctest --output-on-failure -C $BUILD_TYPE from the top-level build directory.

Test results

[doctest] test cases:  56 |  56 passed | 0 failed | 0 skipped
[doctest] assertions: 194 | 194 passed | 0 failed |
[doctest] Status: SUCCESS!

CTest from the top-level build directory:

1/1 Test #1: test_traincascade ................   Passed    0.73 sec
100% tests passed, 0 tests failed out of 1

Coverage

Measured with clang -fprofile-instr-generate -fcoverage-mapping:

Module	Line coverage
o_cvboostparams.cpp, o_cvdtreeparams.cpp	100.00%
o_utils.cpp	93.75%
lbpfeatures.cpp	77.78%
haarfeatures.cpp	70.31%
features.cpp	59.02%
HOGfeatures.cpp	32.12%
imagestorage.cpp	20.13%
boost.cpp	11.16%
cascadeclassifier.cpp	8.09%
o_cvdtree*, o_cvboost*, o_cvcascadeboost*, o_cvstatmodel.cpp	0.00%

The suite achieves near-complete coverage of the public configuration surface (params + factories). The uncovered code is the heavyweight ML training internals reachable only through CvCascadeClassifier::train, which would require integration-style fixtures (real .vec + bg.txt) or further refactoring to be unit-testable.

Verification

• ✅ Builds cleanly under clang 18 with the project's -Werror flags.
• ✅ All 56 test cases / 194 assertions pass locally.
• ✅ ctest discovers and runs the suite from the top-level build dir.
• ✅ No changes to library source or behavior — tests are purely additive.

Out of scope / follow-ups

• Integration tests for CvCascadeClassifier::train using the existing samples in res.
• Round-trip serialization tests (read/write via cv::FileStorage) for *Params types.
• setImage / operator() numerical tests on synthetic images for each feature evaluator.