Rendu: zones NaN transparentes au lieu d'interpolation DTM

L'interpolation fillnodata du DTM générait des artefacts visuels
(zones lissées artificielles). Revenu au DTM avec NaN conservés.

Nouvelle approche pour le rendu:
- NaN remplacés par la médiane des valeurs valides (valeur neutre)
- Après application de la colormap, un masque alpha rend les zones
  NaN transparentes au lieu de les afficher
- interpolation='bilinear' sur imshow pour un rendu lisse
- Figure adaptative: 20-40 pouces selon la taille des données
- DPI 200 pour les images > 3000px

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

lidar_pipeline/rendering.py

@@ -323,13 +323,36 @@ def tif_to_png(tif_file, vis_dir, resolution):
                 valid_data = np.asarray(data.compressed() if hasattr(data, 'compressed') else data.flatten())
                 valid_data = valid_data[~np.isnan(valid_data)]
 
-            # Convert MaskedArray to plain ndarray (NaN-filled) to avoid numpy warnings
-            if isinstance(data, np.ma.MaskedArray):
-                data = np.ma.filled(data, np.nan)
+            # Track NaN mask before converting to plain ndarray
+            nan_mask = None
+            if not is_rgb:
+                if isinstance(data, np.ma.MaskedArray):
+                    nan_mask = data.mask.copy()
+                    data = np.ma.filled(data, np.nan)
+                elif np.any(np.isnan(data)):
+                    nan_mask = np.isnan(data)
+
+                # For rendering: replace NaN with neutral value to avoid interpolation halos
+                if nan_mask is not None and np.any(nan_mask) and len(valid_data) > 0:
+                    fill_value = float(np.median(valid_data))
+                    data[nan_mask] = fill_value
+                    nan_mask = nan_mask  # keep for later
 
             # Apply colormap
             data, cmap, title, legend_label, description, is_rgb_result = _apply_colormap(data, tif_file)
 
+            # Apply NaN mask: make zones without data transparent
+            has_nan_mask = nan_mask is not None and not is_rgb_result
+            if has_nan_mask:
+                # data is normalized 0-1 from _apply_colormap; apply cmap to get RGBA
+                cmap_obj = plt.get_cmap(cmap) if isinstance(cmap, str) else cmap
+                rgba = cmap_obj(data)  # (H, W, 4) float RGBA
+                rgba[nan_mask, 3] = 0.0  # transparent where no data
+                data = rgba
+                is_rgba = True
+            else:
+                is_rgba = False
+
             # Create figure — adapt width to data resolution for sharp rendering
             # At high res (5000+px wide), we need a larger figure to avoid downsampling artifacts
             fig_width = max(20, width / 150)
@@ -342,7 +365,7 @@ def tif_to_png(tif_file, vis_dir, resolution):
                          left=0.06, right=0.88, top=0.93, bottom=0.08)
 
             ax = fig.add_subplot(gs[0])
-            if is_rgb:
+            if is_rgba or is_rgb:
                 im = ax.imshow(data, aspect='equal', origin='upper',
                                interpolation='bilinear')
             else: