Add Python bindings for CUDA cluster finder

- Add bind_ClusterFinderCUDA.hpp with pybind11 bindings for
  ClusterFinderCUDA
- Build CUDA bindings as separate _aare_cuda.so to avoid
  segfaults from mixing nvcc and gcc compiled code in the
  same shared object
- Re-export CUDA classes onto _aare in __init__.py so user
  code uses `from aare import ClusterFinderCUDA` regardless
  of which .so hosts the class
- Factory in ClusterFinder.py selects backend; RuntimeError
  if GPU requested on CPU-only build
- Update python/CMakeLists.txt: _aare_cuda module gated
  behind AARE_CUDA and AARE_PYTHON_BINDINGS
- Add validation notebook: ~20x speedup vs sequential ClusterFinder

python/CMakeLists.txt

@@ -15,20 +15,50 @@ else()
     find_package(pybind11 2.13 REQUIRED)
 endif()
 
-# Add the compiled python extension
-pybind11_add_module(
-    _aare           # name of the module
-    src/module.cpp  # source file
-    )
-
-set_target_properties(_aare PROPERTIES
-    LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}
-)
+# ---- Main CPU module --------------------------------------------------------
+# module.cpp is the only source for the main module. When AARE_CUDA=ON, the
+# CUDA bindings live in a *separate* Python extension (_aare_cuda.so) loaded
+# independently at runtime. This isolates the nvcc-compiled translation unit
+# into its own ELF image so pybind11's type registry cannot be corrupted by
+# weak-symbol collisions between gcc-emitted and nvcc-emitted template
+# instantiations.
+pybind11_add_module(_aare NO_EXTRAS src/module.cpp)
+ 
 target_link_libraries(_aare PRIVATE aare_core aare_compiler_flags)
-
+ 
 target_include_directories(_aare SYSTEM PRIVATE
     $<TARGET_PROPERTY:Minuit2::Minuit2,INTERFACE_INCLUDE_DIRECTORIES>
 )
+ 
+set_target_properties(_aare PROPERTIES
+    LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/aare
+    INTERPROCEDURAL_OPTIMIZATION FALSE
+)
+
+# ---- CUDA module (separate .so) --------------------------------------------
+if(AARE_CUDA)
+    pybind11_add_module(_aare_cuda NO_EXTRAS src/cuda_bindings.cu)
+ 
+    target_link_libraries(_aare_cuda PRIVATE aare_cuda aare_compiler_flags)
+ 
+    target_include_directories(_aare_cuda SYSTEM PRIVATE
+        $<TARGET_PROPERTY:Minuit2::Minuit2,INTERFACE_INCLUDE_DIRECTORIES>
+    )
+ 
+    set_target_properties(_aare_cuda PROPERTIES
+        LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/aare
+        INTERPROCEDURAL_OPTIMIZATION FALSE
+        CUDA_RESOLVE_DEVICE_SYMBOLS ON
+        CUDA_SEPARABLE_COMPILATION ON
+    )
+ 
+    target_compile_options(_aare_cuda PRIVATE
+        $<$<COMPILE_LANGUAGE:CUDA>:--expt-relaxed-constexpr>
+        $<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>
+        $<$<COMPILE_LANGUAGE:CUDA>:-Xcompiler=-fvisibility=hidden>
+        $<$<COMPILE_LANGUAGE:CUDA>:-Xcompiler=-fPIC>
+    )
+endif()
 
 # List of python files to be copied to the build directory
 set( PYTHON_FILES
@@ -51,9 +81,9 @@ foreach(FILE ${PYTHON_FILES})
     configure_file(${FILE} ${CMAKE_BINARY_DIR}/${FILE}  )
 endforeach(FILE ${PYTHON_FILES})
 
-set_target_properties(_aare PROPERTIES
-    LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/aare
-)
+# set_target_properties(_aare PROPERTIES
+#     LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/aare
+# )
 
 set(PYTHON_EXAMPLES
     examples/play.py
@@ -69,8 +99,13 @@ endforeach(FILE ${PYTHON_EXAMPLES})
 
 
 if(AARE_INSTALL_PYTHONEXT)
+    set(AARE_PY_INSTALL_TARGETS _aare)
+    if(AARE_CUDA)
+        list(APPEND AARE_PY_INSTALL_TARGETS _aare_cuda)
+    endif()
+ 
     install(
-        TARGETS _aare
+        TARGETS ${AARE_PY_INSTALL_TARGETS}
         EXPORT "${TARGETS_EXPORT_NAME}"
         LIBRARY DESTINATION aare
         COMPONENT python
@@ -80,5 +115,5 @@ if(AARE_INSTALL_PYTHONEXT)
         FILES ${PYTHON_FILES} 
         DESTINATION aare
         COMPONENT python
-        )
+    )
 endif()