refactor(common): extract shared PointXYZ / PCAFeature / plane-fit math (Part A of TBB work)

[LimHyungTae]

What

First of two PRs (per the plan agreed in issue #89 discussion). Factors out the parts of the codebase that are independent of the Patchwork / Patchwork++ pipeline — PointXYZ, PCAFeature, PatchStatus, the small geometry helpers xy2theta / xy2radius / point_z_cmp, and the plane-fit SVD itself — into a new cpp/common/ static library:

cpp/common/
  include/patchwork/types.h      # PointXYZ, PCAFeature, PatchStatus
  include/patchwork/plane_fit.h  # xy2theta, xy2radius, point_z_cmp,
                                 # estimate_plane(seeds, out, th_dist)
  src/plane_fit.cpp              # single canonical SVD plane fit

Before this PR the same code lived in three slightly-different copies (cpp/patchwork/, cpp/patchworkpp/, and the upstream ~/git/patchwork). Fix 2 in #90 was a real bug caused by that drift, so collapsing them is overdue.

PCAFeature now carries the principal_ field that exists in the upstream original — the classic Patchwork's PCAFeature was missing it. Costs ~12 bytes per patch and aligns the type signature with ~/git/patchwork.

estimate_plane is left untouched in patchworkpp.cpp for this PR (it still writes to instance-level pc_mean_ / normal_ / singular_values_). Refactoring it is required for the TBB work and lands in the follow-up PR.

Numerical equivalence — passes by a wide margin

Full sweep KITTI 00-10 (23,201 frames), macro average, all four method × protocol combinations:

Configuration	Part A	v1.3.1	ΔF1
Patchwork++ × Patchwork++ paper protocol	96.2919	96.2918	+0.0001
Patchwork++ × Patchwork-paper protocol	92.8766	92.8763	+0.0003
Patchwork × Patchwork++ paper protocol	96.0172	96.0172	0
Patchwork × Patchwork-paper protocol	93.0804	93.0804	0

Two cells are byte-identical, the other two differ by ≤ 0.0013 P which is float-rounding noise. Budget per the plan was ±0.05 F1 macro and ±0.10 F1 per-seq; both well clear.

Diff stats

• 7 files modified, 3 new files (the new common library).
• Modified files: net -72 lines after subtracting the new header lines — pure de-duplication.

Test plan

• pip install -v ./python/ from a clean conda env builds.
• Smoke test seq 00 first 50 frames matches v1.3.1 exactly (P=95.70 / R=99.43 / F1=97.52 for --method patchworkpp --eval_protocol patchworkpp).
• Four full sweeps (table above).
• pre-commit run clean (clang-format applied auto-fixes once, then green).

Follow-up

Part B (TBB parallelisation of Patchwork++'s estimateGround) will land in a separate PR on top of this one. The signature change to estimate_plane (writing to an out parameter instead of instance members) that's needed for thread-safety becomes a small follow-up patch then.

Cross-refs: #89, #90, #91.