Draft: Release 5.0

Closes #41, #51, #53, #57

• Update to OTB 9.1
• Move back TF build env and OTB build flags into Dockerfile
• Update bazel build to latest specs (Clang, auto deps download HERMETIC_CUDA)
• Use TF 2.18 + official TF wheels setup (--config release_*)
• Unified build process and base image for CPU and GPU images
• Add build arg CUDA_CC for compute capabilities
• Drop INLINE_CACHE, use dedicated repository cache => reduce images size to half
• Move python installation to venv (required for future build py312)
• Install dev / tests packages in dev image to speedup pipelines
• Run release CI only on tags
• Update core module to Keras3 (example models used in failing tests)
• Move packaging spec to pyproject.toml ( #41 )
• Fix SR4RS test
• Fix decloud test ?
• Update layers module to Keras3 ops ?

otbtf/examples/tensorflow_v2x/fcnn/train_from_patchesimages.py

@@ -90,7 +90,7 @@ def create_dataset(
     tf_ds = dataset.get_tf_dataset(
         batch_size=batch_size,
         preprocessing_fn=fcnn_model.dataset_preprocessing_fn,
-        targets_keys=targets_keys or [fcnn_model.TARGET_NAME]
+        targets_keys=targets_keys or [fcnn_model.OUTPUT_SOFTMAX_NAME]
     )
 
     return tf_ds