Integrate NDT Scan Matching into Awkernel #751

Integrate NDT scan matching into Awkernel using a preloaded point cloud map and externally supplied initial pose estimates. Establish a reproducible baseline before integration with point cloud preprocessing and EKF localization.

https://github.com/tier4/realtime_core/tree/main/realtime_ndt_scan_matcher

Scope:

• Load a fixed point cloud map and construct the NDT map representation before processing scans
• Accept sensor point clouds and an initial pose estimate for each scan
• Define coordinate frames, transformations, timestamps, and required point fields
• Configure NDT resolution, convergence criteria, and iteration limits
• Expose the estimated pose, convergence status, and iteration count
• Measure map preparation separately from per-scan execution time and memory usage

Acceptance criteria:

• NDT scan matching runs on Awkernel with a fixed map and supplied initial pose estimates.
• Estimated poses and convergence outcomes are compared against a fixed reference implementation using identical inputs, initial estimates, and parameters.
• Numerical tolerances and expected behavior are defined before comparison.
• Tests cover representative scans, perturbed initial estimates, empty or invalid inputs, and non-convergence.
• Failed or non-converged results are explicitly reported rather than silently treated as valid estimates.
• Evaluation datasets, configurations, and reproduction steps are documented.

Out of scope:

Crop box filtering, voxel grid downsampling, EKF integration, automatic pose initialization, dynamic map loading, and end-to-end pipeline evaluation.