onnxruntime

12ff718a - docs: update TensorRT RTX EP documentation for standalone EP ABI (#28026)

docs: update TensorRT RTX EP documentation for standalone EP ABI (#28026) ## Summary Updates the TensorRT RTX Execution Provider documentation to reflect the transition to the standalone EP ABI plugin and updated default behavior. ## Changes - **Deprecation notice:** Added a prominent deprecation notice for the built-in TensorRT RTX EP, directing users to the standalone [TensorRT RTX EP ABI](https://github.com/NVIDIA/TensorRT-RTX-EP-ABI) plugin. - **Standalone EP ABI usage instructions:** Added C++ code examples demonstrating the V2 device-based EP API (`RegisterExecutionProviderLibrary`, `GetEpDevices`, `SessionOptionsAppendExecutionProvider_V2`) for registering and using the standalone plugin. - **Build from source:** Split into "Standalone EP ABI Plugin (Recommended)" and "Built-in EP (Deprecated)" sections with appropriate links. - **CUDA Graph default updated:** Updated documentation to reflect that `enable_cuda_graph` is now **enabled by default** (`true`). Code examples and `onnxruntime_perf_test` commands now show how to *disable* CUDA Graph rather than enable it. - **onnxruntime_perf_test usage:** Added EP ABI plugin command-line examples using `--plugin_eps`, `--plugin_ep_libs`, and `--plugin_ep_options` flags alongside the deprecated built-in EP commands for CUDA Graph, EP context generation, and general performance testing sections. - **EP options note:** Clarified that all execution provider options apply to both the built-in EP and the standalone EP ABI plugin. ## Motivation The built-in TensorRT RTX EP is being deprecated in favor of the standalone open-source EP ABI plugin. This documentation update ensures users have clear migration guidance, up-to-date code examples, and accurate default values.