Fix bugs and improve pipeline flexibility

- Fix gpu_cleanup import missing in visualizations.py (NameError in workers)
- Fix t_pdf referenced before assignment when PDF is skipped
- Skip classification+DTM when DTM exists regardless of --force
- --force now only regenerates WebP/PDF, not classification/DTM
- --force-classification forces reclassification when needed
- Add laspy repair fallback for corrupt LAZ files (EVLR errors)
- Keep DTM TIF by default for reuse (--no-keep-tif to delete)
- Increase space between image and bottom cartouche (0.12→0.19)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

lidar_pipeline/dtm.py

@@ -246,10 +246,57 @@ def classify_ground(laz_file, temp_dir, method='auto', force=False):
         logger.info(f"  ✓ Classification sol {method.upper()} terminée")
         return output_las
     except subprocess.CalledProcessError as e:
-        logger.error(f"  ✗ Erreur classification PDAL ({method.upper()}): {e.stderr.decode()}")
+        error_msg = e.stderr.decode() if e.stderr else str(e)
+        logger.warning(f"  ✗ Erreur classification PDAL ({method.upper()}): {error_msg}")
+
+        # Try repairing file with laspy if PDAL fails on EVLR/VLR
+        if 'VLR' in error_msg or 'Invalid' in error_msg:
+            logger.info(f"  → Tentative de réparation du fichier avec laspy...")
+            repaired_las = temp_dir / f"{laz_base}_repaired.las"
+            if _repair_laz_with_laspy(laz_file, repaired_las):
+                # Retry PDAL pipeline with repaired file
+                pipeline_json = _create_ground_pipeline(repaired_las, output_las, method)
+                with open(pipeline_file, 'w') as f:
+                    f.write(pipeline_json)
+                try:
+                    subprocess.run(
+                        ["pdal", "pipeline", str(pipeline_file)],
+                        capture_output=True, check=True
+                    )
+                    logger.info(f"  ✓ Classification sol {method.upper()} terminée (fichier réparé)")
+                    return output_las
+                except subprocess.CalledProcessError as e2:
+                    error_msg2 = e2.stderr.decode() if e2.stderr else str(e2)
+                    logger.error(f"  ✗ Échec classification même après réparation: {error_msg2}")
+            else:
+                logger.error(f"  ✗ Impossible de réparer le fichier")
         return None
 
 
+def _repair_laz_with_laspy(input_laz, output_las):
+    """Try to repair a corrupt LAZ file by re-reading with laspy and saving as LAS.
+
+    Works around PDAL errors like 'Invalid Extended VLR size' by stripping
+    problematic VLR/EVLR metadata during re-save.
+
+    Args:
+        input_laz: Path to corrupt LAZ/LAS file.
+        output_las: Path for repaired LAS output.
+
+    Returns:
+        True if repair succeeded, False otherwise.
+    """
+    import laspy
+    try:
+        las = laspy.read(str(input_laz))
+        las.write(str(output_las))
+        logger.info(f"  ✓ Fichier réparé via laspy ({len(las.points):,} points)")
+        return True
+    except Exception as e:
+        logger.warning(f"  ✗ Réparation laspy échouée: {e}")
+        return False
+
+
 def create_dtm_fast(las_file, basename, dtm_dir, resolution, force=False):
     """Create DTM using fast binning method with gap filling.