Add direct CvDTree / CvDTreeTrainData unit tests

[vladiant]

Summary

Adds 8 direct unit tests for the legacy decision-tree implementation (CvDTree, CvDTreeTrainData) — the two largest uncovered files in TrainCascadeLib. Tests use tiny in-memory cv::Mat datasets and exercise both the high-level train(cv::Mat, ...) entry point and the shared-data overload train(CvDTreeTrainData*, ...) that CvBoostTree / CvCascadeBoostTree use internally.

Motivation

Coverage analysis identified o_cvdtree.cpp (10%) and o_cvdtreetraindata.cpp (cold) as the largest remaining gaps. The cascade trainer drives these classes only indirectly through the boosting layer, so prior end-to-end tests left their core data-prep / split / predict paths largely untouched.

Changes

New file: test_dtree.cpp

Group	Test cases
CvDTree::train(cv::Mat ...)	trains on 1-D separable dataset and builds a root; classifies every training point correctly; generalises to held-out points on each side of the margin; clear() drops the trained tree; re-training on the same instance produces a fresh tree
CvDTree::train(CvDTreeTrainData*)	trains successfully against an externally-built CvDTreeTrainData
CvDTreeTrainData	reports correct is_classifier / sample_count / var_count / get_num_classes() / get_var_type() after set_data; clear() is safe on a default-constructed instance

Helpers (file-local):

• makeOneDimSeparableDataset() — 10 samples, 2 classes, clean margin at x = 0.5.
• makeTinyParams() — max_depth = 3, min_sample_count = 1, no surrogates / CV-folds (required for tiny datasets).
• sample1D(value) — 1×1 CV_32F row for predict().

All tests follow the project's Arrange / Act / Assert convention with comments per step.

Updated: CMakeLists.txt — registers test/test_dtree.cpp in the test_traincascade target.

Verification

[doctest] test cases:  96 |  96 passed | 0 failed | 0 skipped
[doctest] assertions: 321 | 321 passed | 0 failed |
[doctest] Status: SUCCESS!

Test-suite runtime unchanged (~17 s).

Coverage impact

Metric	Before	After	Δ
Lines	44.4% (2077 / 4676)	57.5% (2687 / 4676)	+13.1 pp
Functions	79.1% (204 / 258)	86.4% (223 / 258)	+7.3 pp
Branches	28.8% (1443 / 5006)	36.8% (1840 / 5006)	+8.0 pp

Single largest coverage uplift in the test-suite history — activates ~610 previously-uncovered lines across o_cvdtree.cpp and o_cvdtreetraindata.cpp.

Risk

Test-only change. No production source modified.

Checklist

• All 96 tests pass locally (clang 18, Linux, OpenCV 4.6).
• Coverage verified with gcovr against the GCC 13 instrumented build.
• No new compiler warnings.
• Follows existing AAA-with-comments doctest pattern used elsewhere in test.