PiC 3.3.1

Changes

Moved lidR from Imports to Suggests: LAS/LAZ reading and writing are now optional and guarded with requireNamespace(), so the package installs and works (with .xyz/.txt data) even when lidR is unavailable.
Declared the maintainer repository for lidR via Additional_repositories.

Substantially expanded the Forest_seg() documentation: full 8-stage pipeline, point classification scheme (forest floor, understory, wood, crown, non-valid trees, noise), the LOR, DAV and GAB algorithms, DBH measurement, the LAS/XYZ outputs, the tree/plot reports and the parameters log, and the return value. Also noted that Forest_seg() is designed for very large clouds (up to ~600M points / >10 GB on a 36 GB workstation, ~1M points/s).

New function metrics_from_las(): recomputes tree- and plot-level metrics (DBH, height, crown base height, plot statistics) from an already classified point cloud (LAS with ASPRS classes ground = 2, understory = 3, wood = 4, crown = 5), without repeating the full Forest_seg() segmentation.
New function pic_analyze_cloud(): pre-processing diagnostic tool that assesses point-cloud quality and structure (point density per m^2, voxel density distribution), returns plots for interactive use and optionally exports a PDF report. The loaded cloud stays in memory for a subsequent Forest_seg() run.
Forest_seg() plot report now includes crown_volume_m3 and understory_volume_m3, computed from the final classified points (count of occupied voxels at canopy_vox_dim resolution times the voxel volume) and rounded to whole cubic metres.

Crown base height (GAB): branch connectivity now uses direct vertical adjacency only (no gap-layer bridging). The BFS keeps the six lateral hexagonal neighbours, but a single empty layer now breaks a branch instead of being bridged. CBH therefore reflects the lowest foliated branch rather than bridging into the upper crown. cbh_max_gap_layers is retained for backward compatibility but is no longer used.
Rescue pass (LOR) revised:
- also considers pure-noise voxels (cls == 0), not only rejected clusters;
- h_trunk_factor is clamped to [0.3, 1.0], so rescue is never stricter than the base pass on height;
- DBSCAN minPts is halved, so sparse noise can re-cluster;
- a stricter linearity threshold than the base pass is applied (low trunk-band segments must be more linear), to avoid admitting bushes.
output_format: LAS is the default; any value other than "xyz" falls back to LAS.
Report CSVs keep the dot (.) as decimal separator for international use; crown/understory volumes are written as whole cubic metres.
Removed the Crown Packing Index (CPI) and the unused internal helper .analyze_canopy_float(), which was never wired into the pipeline.
Removed the unused internal helper .assign_components_bfs_float().

Fixed the voxel-to-point reclassification (DAV) that could fail when the DAV step returned early without an H_threshold value; it now falls back to dav_understory_max_start.
Documented parameter defaults now match the function signature.
Internal helper .lor_create_voxels() is marked internal and no longer appears in the package help index.

Updated defaults for Forest_seg():