Remove PMF, fix NaN in gradient visualizations, fix pos_open/neg_open shared param

- Remove PMF from ground classification options (PDAL recommends SMRF over PMF)
- Auto-detection now uses CSF for urban/complex terrain instead of PMF
- Add z_std > 30m heuristic to auto-select CSF for complex terrain
- Fix pos_open/neg_open lambda missing 'shared' parameter (NameError in workers)
- Fix NaN mask not restored in hillshade, slope, aspect, curvature
  (gradient-based products computed on filled DEM lost NaN transparency)
- Add nan_mask parameter to _save_tif for centralized NaN restoration
- DTM TIF kept by default (no longer deleted after WebP conversion)

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

lidar_pipeline/dtm.py

@@ -1,6 +1,6 @@
 """DTM generation from classified LiDAR point clouds.
 
-Handles ground classification via PDAL/SMRF and DTM rasterisation
+Handles ground classification via PDAL (SMRF or CSF) and DTM rasterisation
 using scipy binned_statistic_2d. Zones without LiDAR data remain as NaN.
 """
 
@@ -27,13 +27,13 @@ def _create_ground_pipeline(input_laz, output_las, method):
     1. Reset Classification to 0
     2. ELM (Extended Local Minimum) — mark low outliers as noise (Classification=7)
     3. Statistical outlier removal
-    4. Ground classification (SMRF/PMF/CSF)
+    4. Ground classification (SMRF or CSF)
     5. Extract ground points (Classification=2)
 
     Args:
         input_laz: Path to input LAZ/LAS file.
         output_las: Path to output classified LAS file.
-        method: Ground classification method ('smrf', 'pmf', or 'csf').
+        method: Ground classification method ('smrf' or 'csf').
 
     Returns:
         JSON string of the PDAL pipeline.
@@ -84,15 +84,6 @@ def _create_ground_pipeline(input_laz, output_las, method):
             "threshold": 0.5,
             "scalar": 1.25
         }
-    elif method == 'pmf':
-        ground_step = {
-            "type": "filters.pmf",
-            "max_window": 33,
-            "slope": 0.15,
-            "max_distance": 2.5,
-            "initial_distance": 0.15,
-            "cell_size": 1.0
-        }
     elif method == 'csf':
         ground_step = {
             "type": "filters.csf",
@@ -128,11 +119,6 @@ def create_smrf_pipeline(input_laz, output_las):
     return _create_ground_pipeline(input_laz, output_las, 'smrf')
 
 
-def create_pmf_pipeline(input_laz, output_las):
-    """Create a PDAL pipeline JSON for PMF ground classification."""
-    return _create_ground_pipeline(input_laz, output_las, 'pmf')
-
-
 def create_csf_pipeline(input_laz, output_las):
     """Create a PDAL pipeline JSON for CSF ground classification."""
     return _create_ground_pipeline(input_laz, output_las, 'csf')
@@ -141,9 +127,9 @@ def create_csf_pipeline(input_laz, output_las):
 def detect_ground_method(laz_file):
     """Detect the best ground classification method based on point cloud statistics.
 
-    Auto-selects between SMRF (natural terrain) and PMF (urban) only.
-    CSF is available only via --ground-classification csf (slow but robust
-    on complex terrain).
+    Auto-selects between SMRF and CSF:
+    - SMRF: fast, robust for most natural terrain (PDAL recommended default)
+    - CSF: cloth simulation, better for complex/urban terrain
 
     Falls back to SMRF if the file cannot be read or attributes are missing.
 
@@ -151,7 +137,7 @@ def detect_ground_method(laz_file):
         laz_file: Path to input LAZ/LAS file.
 
     Returns:
-        String: 'smrf', 'pmf', or 'csf'
+        String: 'smrf' or 'csf'
     """
     import laspy
 
@@ -182,13 +168,16 @@ def detect_ground_method(laz_file):
                 f"ratio_retours_uniques={single_return_ratio:.2f}, "
                 f"écart_Z={z_std:.1f}m, amplitude_Z={z_range:.1f}m")
 
-    # Decision logic (auto selects between SMRF and PMF only):
-    # - High single-return ratio (>0.6) → urban (buildings, roads) → PMF
+    # Decision logic:
+    # - High single-return ratio (>0.6) → urban (buildings, roads) → CSF (cloth simulation)
+    # - High elevation variance (>30m) → complex/mountainous terrain → CSF
     # - Default → SMRF (fast, robust for most natural terrain)
-    # Note: CSF is available only via --ground-classification csf (slow but robust on complex terrain)
     if single_return_ratio > 0.6:
-        method = 'pmf'
+        method = 'csf'
         reason = f"ratio retours uniques={single_return_ratio:.2f} > 0.6 → milieu urbain"
+    elif z_std > 30:
+        method = 'csf'
+        reason = f"écart_Z={z_std:.1f}m > 30m → terrain complexe"
     else:
         method = 'smrf'
         reason = f"terrain naturel standard"
@@ -203,7 +192,7 @@ def classify_ground(laz_file, temp_dir, method='auto', force=False):
     Args:
         laz_file: Path to input LAZ/LAS file.
         temp_dir: Directory for temporary files (pipeline.json, ground.las).
-        method: Ground classification method ('auto', 'smrf', 'pmf', or 'csf').
+        method: Ground classification method ('auto', 'smrf', or 'csf').
         force: If True, reclassify even if output file already exists.
 
     Returns: