Add CUDA custom op header files to Linux tarball (#21551)

### Description
The header files were added in PR #16454. 
Then, recently I made a PR #21464 that changed how we packed Linux
tarballs.
The new tarball misses the custom op header files.
Therefore I need to make this change.


### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->

Author: snnn

cmake/onnxruntime.cmake

@@ -38,10 +38,14 @@ function(get_c_cxx_api_headers HEADERS_VAR)
 
   # need to add header files for enabled EPs
   foreach(f ${ONNXRUNTIME_PROVIDER_NAMES})
-    file(GLOB _provider_headers CONFIGURE_DEPENDS
-      "${REPO_ROOT}/include/onnxruntime/core/providers/${f}/*.h"
-    )
-    list(APPEND _headers ${_provider_headers})
+    # The header files in include/onnxruntime/core/providers/cuda directory cannot be flattened to the same directory 
+    # with onnxruntime_c_api.h . Most other EPs probably also do not work in this way.
+    if((NOT f STREQUAL cuda) AND (NOT f STREQUAL rocm))
+      file(GLOB _provider_headers CONFIGURE_DEPENDS
+        "${REPO_ROOT}/include/onnxruntime/core/providers/${f}/*.h"
+      )
+      list(APPEND _headers ${_provider_headers})
+    endif()
   endforeach()
 
   set(${HEADERS_VAR} ${_headers} PARENT_SCOPE)

cmake/onnxruntime_providers_cpu.cmake

@@ -219,6 +219,7 @@ if (onnxruntime_ENABLE_TRAINING)
 endif()
 
 install(FILES ${PROJECT_SOURCE_DIR}/../include/onnxruntime/core/providers/cpu/cpu_provider_factory.h  DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/onnxruntime/)
+install(FILES ${PROJECT_SOURCE_DIR}/../include/onnxruntime/core/providers/resource.h ${PROJECT_SOURCE_DIR}/../include/onnxruntime/core/providers/custom_op_context.h DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/onnxruntime/core/providers)
 set_target_properties(onnxruntime_providers PROPERTIES LINKER_LANGUAGE CXX)
 set_target_properties(onnxruntime_providers PROPERTIES FOLDER "ONNXRuntime")
 

cmake/onnxruntime_providers_cuda.cmake

@@ -289,8 +289,15 @@
     config_cuda_provider_shared_module(onnxruntime_providers_cuda_obj)
   endif()
   config_cuda_provider_shared_module(onnxruntime_providers_cuda)
-
+  # Cannot use glob because the file cuda_provider_options.h should not be exposed out.
+  set(ONNXRUNTIME_CUDA_PROVIDER_PUBLIC_HEADERS
+        "${REPO_ROOT}/include/onnxruntime/core/providers/cuda/cuda_context.h"
+        "${REPO_ROOT}/include/onnxruntime/core/providers/cuda/cuda_resource.h"
+      )
+  set_target_properties(onnxruntime_providers_cuda PROPERTIES
+    PUBLIC_HEADER "${ONNXRUNTIME_CUDA_PROVIDER_PUBLIC_HEADERS}")
   install(TARGETS onnxruntime_providers_cuda
+          PUBLIC_HEADER DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/onnxruntime/core/providers/cuda
           ARCHIVE  DESTINATION ${CMAKE_INSTALL_LIBDIR}
           LIBRARY  DESTINATION ${CMAKE_INSTALL_LIBDIR}
           RUNTIME  DESTINATION ${CMAKE_INSTALL_BINDIR})

cmake/onnxruntime_providers_rocm.cmake

@@ -223,8 +223,13 @@
   if (onnxruntime_ENABLE_ATEN)
     target_compile_definitions(onnxruntime_providers_rocm PRIVATE ENABLE_ATEN)
   endif()
-
+  file(GLOB ONNXRUNTIME_ROCM_PROVIDER_PUBLIC_HEADERS CONFIGURE_DEPENDS
+        "${REPO_ROOT}/include/onnxruntime/core/providers/rocm/*.h"
+      )
+  set_target_properties(onnxruntime_providers_rocm PROPERTIES
+    PUBLIC_HEADER "${ONNXRUNTIME_ROCM_PROVIDER_PUBLIC_HEADERS}")
   install(TARGETS onnxruntime_providers_rocm
+          PUBLIC_HEADER DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/onnxruntime/core/providers/rocm
           ARCHIVE  DESTINATION ${CMAKE_INSTALL_LIBDIR}
           LIBRARY  DESTINATION ${CMAKE_INSTALL_LIBDIR}
           RUNTIME  DESTINATION ${CMAKE_INSTALL_BINDIR})