Infrastructure for Machine Learning inference with ONNX Runtime

This MR regroups the existing OnnxRuntime packages into an Onnx-specific package located at Control/AthOnnx. Breakdown of the changes.

• The new Onnx package implementation can be found in commit 61a8f24f. Accordingly, the ONNX example and GNN tracking are updated in the commit c3bf27d9.

• A small change in Control/AthenaConfiguration/python/AllConfigFlags.py to include the ONNX flags so that users can easily configure Onnx backends: here

• Remove the two packages that are no longer needed: Control/AthOnnxruntimeService and Control/AthOnnxruntimeUtils, see 0ac0c9a2. Because a couple of other packages are using the Control/AthOnnxruntimeService, keep it for now and update the Onnx inference in another MR.

Adding @akraszna @dbakshig

Control/AthOnnx/AthOnnxComps/python/OnnxRuntimeFlags.py

+# Copyright (C) 2002-2024 CERN for the benefit of the ATLAS collaboration
+
+from AthenaConfiguration.AthConfigFlags import AthConfigFlags
+from AthenaConfiguration.Enums import FlagEnum
+
+class OnnxRuntimeType(FlagEnum):
+    CPU = 'CPU'
+    CUDA = 'CUDA'
+# possible future backends. Uncomment when implemented.
+    # DML = 'DML'
+    # DNNL = 'DNNL'
+    # NUPHAR = 'NUPHAR'
+    # OPENVINO = 'OPENVINO'
+    # ROCM = 'ROCM'
+    # TENSORRT = 'TENSORRT'
+    # VITISAI = 'VITISAI'
+    # VULKAN = 'VULKAN'
+
+def createOnnxRuntimeFlags():
+    icf = AthConfigFlags()
+
+    icf.addFlag("AthOnnx.ExecutionProvider", OnnxRuntimeType.CPU, type=OnnxRuntimeType)
+
+    return icf
+
+if __name__ == "__main__":
+
+    flags = createOnnxRuntimeFlags()
+    flags.dump()