## Defensive Coding Improvements: batch_processor.py & feature_extractor.p

[agentksimha]

Summary

Adds input validation, output safety guards, and fixes silent failure points across the PLY processing pipeline.

---

batch_processor.py

[High] Path & permission validation (_validate_path)
Introduces a _validate_path helper called at the top of validate_ply_file. Checks that the file exists, is a regular file, carries a .ply extension, and is readable before Open3D is ever invoked — replacing opaque internal errors with clear, early messages.

[High] File size validation (_check_file_size)
Rejects files under 80 bytes (corrupt/empty) as a hard error and warns on files exceeding 500 MB before any processing begins.

[Medium] Output path safety (preprocess_file)
Aborts if the resolved output path matches the input (preventing silent data loss) and verifies the output directory is writable before any mesh work is done.

[Medium] Zero-extent guard (normalize_mesh_scale)
Prevents a division-by-zero crash on degenerate meshes with zero bounding-box extent. Also validates that target_size is a positive number.

---

feature_extractor.py

[Refactor] PCA swap in compute_geometric_moments
Replaces manual np.cov + np.linalg.eigh with sklearn.decomposition.PCA, consistent with compute_surface_normal which already used PCA. No behaviour change.

[Medium] No in-place mutation (compute_surface_curvature)
Deep-copies the input point cloud before calling estimate_normals(), preventing silent permanent modification of the caller's object.

[Medium] JSON-serialisable boundary points (compute_boundary_features)
Converts boundary_points from a raw numpy array to a plain Python list, consistent with all other fields returned by this class and safe for json.dumps(). Tightens bare except to except Exception as e and surfaces the error string.

[Medium] KeyError-safe visualisation (visualize_surface_features)
Extends the existing surface guard to also check that features have been extracted before accessing fragment['features'][color][surface_idx], with a distinct message for each failure case. Adds a np.array() cast for boundary_points at the plot site since it is now stored as a list.