Add multi-GPU support and fix scale bar / location map overlap

Multi-GPU:
- gpu.py: lazy CuPy initialization so CUDA_VISIBLE_DEVICES takes
  effect before context creation in worker processes
- gpu.py: detect GPU count via nvidia-smi (no CUDA import needed)
- gpu.py: add set_active_gpu() to assign workers to specific GPUs
- pipeline.py: distribute files across GPUs (file % num_gpus) in
  parallel mode so both GPUs are used simultaneously
- pipeline.py: log GPU count when multiple GPUs detected

Layout fixes:
- rendering.py: move scale bar left of location map to avoid overlap
  (scale bar ends at fig_x=0.78, map starts at 0.82)
- rendering.py: expand location map inset to 0.16x0.13 fig coords
- rendering.py: return bounds from _download_location_map so imshow
  extent matches the actual IGN tile coverage (80km context)
- ign.py: add min_zoom parameter to download_ign_tiles, fixing the
  location map that was broken (zoom 10 blocked by hardcoded min_zoom=15)

lidar_pipeline/rendering.py

@@ -274,15 +274,16 @@ def _apply_colormap(data, tif_file):
 def _download_location_map(min_x, max_x, min_y, max_y):
     """Download a wide-area IGN topographic map for location context.
 
-    Downloads a zoomed-out IGN PLANIGNV2 tile covering 5-10x the processed
-    zone extent, giving a wider geographic context. Results are cached to
+    Downloads a zoomed-out IGN PLANIGNV2 tile covering ~80km around the
+    processed zone, giving wider geographic context. Results are cached to
     avoid re-downloading for each visualization in the same tile.
 
     Args:
         min_x, max_x, min_y, max_y: DTM bounds in Lambert 93.
 
     Returns:
-        numpy array (H, W, 3) uint8, or None on failure.
+        Tuple (image_array, bounds_dict) where bounds_dict has keys
+        'min_x', 'max_x', 'min_y', 'max_y' in Lambert 93, or None on failure.
     """
     import hashlib
 
@@ -323,7 +324,13 @@ def _download_location_map(min_x, max_x, min_y, max_y):
         )
 
         if result is not None:
-            _location_map_cache[cache_key] = result
+            bounds = {
+                'min_x': context_min_x, 'max_x': context_max_x,
+                'min_y': context_min_y, 'max_y': context_max_y,
+            }
+            cached = (result, bounds)
+            _location_map_cache[cache_key] = cached
+            return cached
 
         return result
     except Exception as e:
@@ -637,6 +644,7 @@ def tif_to_png(tif_file, vis_dir, resolution, keep_tif=False, source_info=None,
                                   edgecolor='#cccccc', alpha=0.9))
 
             # Scale bar — adaptive with alternating black/white segments
+            # Position: left of the location map to avoid overlap
             extent_m_x = max_x - min_x
             scale_m, scale_label = _nice_scale(extent_m_x)
             pixels_per_meter = 1.0 / pixel_size_x
@@ -646,7 +654,17 @@ def tif_to_png(tif_file, vis_dir, resolution, keep_tif=False, source_info=None,
             bar_bottom_y = 0.55
             bar_top_y = 0.85
             bar_height = bar_top_y - bar_bottom_y
-            scale_start_x = 0.80
+            # Place scale bar so it ends before the location map (map starts at x=0.82 in fig coords)
+            # map_ax occupies [0.82, 0.02, 0.16, 0.13] in figure coords
+            # info_ax occupies [data_left, 0.015, width, 0.09]
+            # Scale bar end in fig coords = info_ax.left + (scale_start_x + scale_px/width) * info_ax.width
+            # We need: info_ax.left + (scale_start_x + scale_px/width) * info_ax.width < 0.80
+            scale_end_frac = scale_px / width  # fraction of info_ax width
+            info_ax_width = data_width_frac + cbar_width + 0.02
+            # Calculate scale_start_x so scale bar ends at fig_x = 0.78 (leaving gap before map at 0.82)
+            max_scale_end_fig = 0.78
+            scale_end_in_info = (max_scale_end_fig - data_left) / info_ax_width
+            scale_start_x = max(0.05, scale_end_in_info - scale_end_frac)
 
             for seg_i in range(n_segments):
                 color = 'black' if seg_i % 2 == 0 else 'white'
@@ -666,15 +684,17 @@ def tif_to_png(tif_file, vis_dir, resolution, keep_tif=False, source_info=None,
                         color='black', linewidth=1, transform=info_ax.transAxes, clip_on=False)
 
             # Location inset map — IGN topographic background with processed zone marker
-            map_ax = fig.add_axes([0.84, 0.02, 0.14, 0.12])
+            # Positioned in lower-right corner, above the info bar
+            map_ax = fig.add_axes([0.82, 0.02, 0.16, 0.13])
 
             # Try to download a wide-area IGN topo map for location context
-            location_map = _download_location_map(min_x, max_x, min_y, max_y)
-            if location_map is not None:
-                # Draw IGN topo map as background
+            location_result = _download_location_map(min_x, max_x, min_y, max_y)
+            if location_result is not None:
+                location_map, loc_bounds = location_result
+                # Draw IGN topo map as background with correct bounds
                 map_ax.imshow(location_map, aspect='equal', extent=[
-                    min_x - (max_x - min_x) * 2, max_x + (max_x - min_x) * 2,
-                    min_y - (max_y - min_y) * 2, max_y + (max_y - min_y) * 2
+                    loc_bounds['min_x'], loc_bounds['max_x'],
+                    loc_bounds['min_y'], loc_bounds['max_y']
                 ])
                 # Mark the processed zone with a red rectangle
                 rect_x1, rect_x2 = min_x, max_x