Extend CvDTree branch coverage (regression / 2-D / pruning / sampleIdx)

[vladiant]

Summary

Adds 4 direct unit tests for CvDTree::train that drive code paths the cascade trainer never touches (it always runs depth-1 classification without surrogates, regression, or CV folds). All tests follow the project's Arrange / Act / Assert convention with comments per step.

Changes

Modified: test_dtree.cpp

#	Test case	Drives
1	regression mode learns a step-function on a 1-D dataset	varType = (ORDERED, ORDERED) + regression_accuracy = 0.01 → find_split_ord_reg, regression branch of calc_node_value
2	2-D classification picks an axis-aligned split	2 ordered features + 1 categorical response → multi-variable scan in find_best_split (DTreeBestSplitFinder)
3	cv_folds pruning produces a usable tree on a noisy dataset	60 samples, cv_folds = 3, use_1se_rule = true, truncate_pruned_tree = true → prune_cv / free_prune_data / fold-allocation paths
4	sampleIdx mask trains on the selected subset only	8U mask of length sample_count (selects 6 / 10 rows) → sample-mask preprocessing in CvDTreeTrainData::set_data

No production source modified.

Verification

[doctest] test cases: 107 | 107 passed | 0 failed | 0 skipped
[doctest] assertions: 360 | 360 passed | 0 failed |
[doctest] Status: SUCCESS!

Coverage impact

Metric	Before	After	Δ
Lines	57.7% (2697 / 4676)	62.3% (2912 / 4676)	+4.6 pp
Functions	86.4% (223 / 258)	88.0% (227 / 258)	+1.6 pp
Branches	37.2% (1857 / 5006)	40.7% (2037 / 5006)	+3.5 pp

Net gain: +215 lines and +180 branches covered, concentrated in the previously cold legacy CART machinery:

File	Before	After
o_cvdtree.cpp	~34%	(improved)
o_cvdtreetraindata.cpp	~57%	(improved)
o_cvboostree.cpp	~41%	(improved)
o_cvcascadeboosttree.cpp	~51%	(improved)

(Per-file deltas: regression test alone exercises the regression branch of calc_node_value and find_split_ord_reg; the cv_folds test exercises the prune / fold-allocation paths; the sampleIdx mask test exercises the sample-preprocessing branch in set_data.)

Risk

Test-only change. No production source modified.

Notes

• CvDTreeTrainData rejects cv_folds when per-fold sample count is too low — pruning test uses 60 samples / 3 folds for that reason (documented inline).
• sampleIdx test uses an 8U mask rather than a 32S index array; both paths were considered, the mask form additionally walks the mask-expansion branch in set_data.

Checklist

• All 107 tests pass locally (clang 18 / Linux / OpenCV 4.6).
• Coverage verified with gcovr against the GCC 13 instrumented build.
• No new compiler warnings.
• AAA-with-comments doctest pattern followed throughout.