[QNN EP] Add Support for If Op using Graph Transformation

[quic-ashwshan]

• Currently Onnx graph transform handles only static If Op cases.
  1. When If op input is static and both branches are also static
  2. When if Op input is statifc but branches are dynamic.
• Extend If Op handling to support dynamic condition and dynamic branches.
• Lower both subgraphs to the main graph and select output using ONNX Where op.
• Added unit tests in graph transformer to test dynamic if conditions and ensure that a Where Op is added

Description

This change adds support for lowering ONNX If operations where both the condition and the branches are dynamic. Previously, the ORT transformation logic handled only two cases:

1. Fully static If nodes were folded to constants.
2. Semi-dynamic If nodes (static condition, dynamic branches) were partially lowered by selecting the active branch at compile time.

This change now includes the case -
When both the condition and branches are dynamic, both subgraphs are lowered into the main graph, and a Where op is used to select the correct output at runtime.

Motivation and Context

1. Enables support for dynamic control flow in ONNX models where the condition and both branches of an If node are not statically known.
2. This also helps offload the op to the NPU for QNN backend.

[HectorSVC]

/azp run Linux QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows ARM64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline,Windows x64 QNN CI Pipeline

[azure-pipelines]

Azure Pipelines successfully started running 5 pipeline(s).