Feature/bench kdtree flann vs nanoflann

[rtkartista]

Closes #3860 (partial — search module benchmark)

What

Adds the first benchmark to the search module: benchmarks/search/bench_kdtree.cpp

Benchmarks pcl::search::KdTree (FLANN) vs pcl::search::KdTreeNanoflann
across three axes:

• Tree construction: 10k / 100k / 500k points
• kNN query: k = 1, 5, 20, 50
• Radius search: r = 5, 10, 20 m

Results

Hardware: 12 threads / Ubuntu 24.04 / GCC 13 / -O3 -mavx2
Cloud: table_scene_mug_stereo_textured.pcd (307,200 pts) for query benchmarks;
synthetic uniform random for build benchmarks
Repetitions: 3 — mean reported

absolute timings may vary slightly

Tree construction (ms, lower is better)

Points	FLANN (ms)	nanoflann (ms)	Speedup
10k	1.15	0.979	1.17×
100k	16.2	13.1	1.24×
500k	129	94.4	1.37×

kNN query (µs, lower is better)

k	FLANN (µs)	nanoflann (µs)	Speedup
1	0.831	0.244	3.41×
5	1.43	0.466	3.07×
20	2.18	0.912	2.39×
50	5.10	2.97	1.72×

Radius search (µs, lower is better)

radius (m)	FLANN (µs)	nanoflann (µs)	Speedup
5	12,735	1,521	8.37×
10	12,991	1,536	8.46×
20	13,244	1,539	8.60×

nanoflann is 1.2–1.4× faster to build, 1.7–3.4× faster on kNN, and
8.4× faster on radius search — making it the clear default for
localization and mapping pipelines that call radiusSearch in tight loops
(normal estimation, ICP, GICP, NDT).

Notes

• Accepts optional PCD path as argv[1]; falls back to synthetic 100k-pt cloud
• nanoflann benchmarks compiled in automatically when PCL_HAS_NANOFLANN is defined
• Ready to extend with registration / features benchmarks in follow-up PRs