Layout uniforme WebP: axes fixes + aspect='equal' pour superposition géolocalisée

- Positions d'axes fixes (data_left/bottom/width/height_frac) pour alignement
  pixel-parfait entre terrain et ortho/topo
- aspect='equal' au lieu de 'auto' pour conserver les proportions géographiques
- Colorbar descriptive pour les visualisations RGB (ortho/topo)
- Comblage des petits trous DTM (< 1m) via rasterio.fill.fillnodata
- Suppression de la visualisation "dépressions"
- Hillshade composite: 0.7*hillshade + 0.3*cos(slope)
- D8 flow accumulation accéléré par numba JIT (fallback Python)
- Flag --keep-tif pour conserver les TIFF intermédiaires
- --force supprime aussi les TIF existants avant régénération
- ETA affiché pendant la génération des visualisations
- Répertoires temp dans temp/ pour traitement parallèle

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

CLAUDE.md

@@ -4,7 +4,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 ## Project Overview
 
-LiDAR archaeological processing pipeline that generates 18 terrain visualizations from LAZ/LAS point clouds. Runs in Docker with optional NVIDIA GPU acceleration (CuPy). Designed for French LiDAR HD data in Lambert 93 (EPSG:2154).
+LiDAR archaeological processing pipeline that generates 17 terrain visualizations from LAZ/LAS point clouds. Runs in Docker with optional NVIDIA GPU acceleration (CuPy). Designed for French LiDAR HD data in Lambert 93 (EPSG:2154).
 
 ## Commands
 
@@ -17,6 +17,7 @@ All commands run inside Docker. Use `./run.sh` as the primary interface.
 ./run.sh --test                                 # Run unit tests
 ./run.sh -g --file LHD_FXX_1000_6882_PTS_LAMB93_IGN69.copc  # Single file
 ./run.sh --ground-classification pmf            # Force PMF ground classification
+./run.sh -g --keep-tif                          # Keep intermediate TIFF files
 ./run.sh                                        # Print help (no args)
 ```
 
@@ -33,7 +34,7 @@ docker run --rm --gpus all -v $(pwd)/input:/data/input:ro -v $(pwd)/output:/data
 - **`cli.py`** — argparse + logging setup. Entry point via `python -m lidar_pipeline`.
 - **`pipeline.py`** — `LidarArchaeoPipeline` orchestrator. `VIZ_STEPS` registry maps names to generate functions. `FilePrefixFilter` for parallel logging.
 - **`dtm.py`** — PDAL ground classification (SMRF/PMF/CSF + auto-detection) and DTM generation via scipy `binned_statistic_2d`.
-- **`visualizations.py`** — 16 `generate_*` functions + 2 IGN overlay lambdas. All take `(dem_file, basename, vis_dir, resolution)` and return a TIF path or None.
+- **`visualizations.py`** — 15 `generate_*` functions + 2 IGN overlay lambdas. All take `(dem_file, basename, vis_dir, resolution)` and return a TIF path or None.
 - **`gpu.py`** — CuPy/numpy abstraction: `HAS_GPU`, `to_gpu()`, `to_cpu()`, `xp_gaussian_filter()`, `xp_uniform_filter()`, `xp_minimum_filter()`, `gpu_cleanup()`. Falls back to CPU gracefully.
 - **`ign.py`** — IGN WMTS tile download + overlay generation for orthophoto and topographic maps.
 - **`rendering.py`** — `COLORMAPS` dict maps filename keywords to (cmap, title, legend, description). `tif_to_png()` converts TIF→WebP with legend/scale/north arrow. `generate_pdf_report()` creates A3 PDF.
@@ -56,7 +57,7 @@ Override with `--ground-classification {auto,smrf,pmf,csf}`.
 
 ### NaN handling
 
-DTM zones without LiDAR data are kept as NaN (no interpolation). Visualization functions use `_fill_nans()` and `_filter_nanaware()` to avoid NaN propagation through filters.
+DTM small gaps (< 1m from existing data) are filled using `rasterio.fill.fillnodata`. Large gaps remain as NaN. Visualization functions use `_fill_nans()` and `_filter_nanaware()` to avoid NaN propagation through filters.
 
 ### Parallel processing
 
@@ -67,5 +68,6 @@ Uses `ProcessPoolExecutor` with `'spawn'` start method (required for CUDA). Each
 - **Language**: UI messages and comments in French. Code identifiers in English.
 - **Logging**: Use `logger = logging.getLogger("lidar")`. Prefix per-file logs via `_file_filter.basename`.
 - **GPU pattern**: `arr_gpu = to_gpu(arr)` → compute → `result = to_cpu(arr_gpu)` → `gpu_cleanup()` between visualizations.
-- **Output format**: Visualizations saved as WebP (not PNG). PDF reports use `PILImage.open().convert('RGB')`.
+- **Output format**: Visualizations saved as WebP (not PNG). TIFF intermediates deleted unless `--keep-tif`. PDF reports use `PILImage.open().convert('RGB')`.
+- **Flow accumulation**: Uses numba JIT for D8 accumulation loop. Falls back to pure Python if numba unavailable.
 - **Tests**: Run only inside Docker via `./run.sh --test`. Synthetic DEM fixture in `tests/conftest.py`.