Support pre-registered plug-in NvTensorRtRtx execution provider library

[anujj]

This PR basically extend the support for pre-registered plug-in for NvTenosrRtRTX which was earlier done for Cuda in the PR: #1850

What We Are Doing

• Enable NvTensorRtRtx pre-registered plugin provider support - Allow users to register custom NvTensorRT execution provider libraries (similar to existing CUDA plugin support)
• Add example code demonstrating plugin provider usage in C++ and Python

Implementation Flow
Pre-Registered Plugin Path (PATH A)
When a custom EP library is registered

    app Code:
      └─> register_execution_provider_library("NvTensorRTRTXExecutionProvider", "/path/to/*dll")
           └─> Creates model with provider="NvTensorRtRtx"
    
    Model Creation (model.cpp):
      └─> Check GetOrtEnv().GetEpDevices() for pre-registered "NvTensorRTRTXExecutionProvider"
           └─> IF FOUND (PATH A):
                └─> Use AppendExecutionProvider_V2() with registered provider

Built-in/Default Path (PATH B)
When no plugin is registered (standard behavior):

User Code:
  └─> Creates model with provider="NvTensorRtRtx" (no pre-registration)

  Model Creation (model.cpp):
    └─> Check GetOrtEnv().GetEpDevices() for pre-registered "NvTensorRTRTXExecutionProvider"
         └─> IF NOT FOUND (PATH B):
              ├─> FALL THROUGH to generic provider registration block
              └─> Generic code uses AppendExecutionProvider() with built-in EP
                   └─> On WinML: includes device filtering (multi-GPU support)
                   └─> On non-WinML: simple provider append

[anujj]

@gaugarg-nv @ishwar-raut1 for review

[anujj]

Adding @kunal-vaishnavi and @baijumeswani for review

[kunal-vaishnavi]

Can we generalize to something like this?

bool already_handled = std::any_of(GetOrtEnv().GetEpDevices().begin(),
                                   GetOrtEnv().GetEpDevices().end(),
                                   [](const auto& device_ptr) { return device_ptr->Name() == "NvTensorRTRTXExecutionProvider"; });
if (already_handled) return;

// For providers that go through the extensible AppendExecutionProvider API

[anujj]

done

[anujj]

@kunal-vaishnavi : I have addressed the review comments in the latest commit