feat: add TRT-RTX native CUDA graph support

[tp5uiuc]

Description

Add cuda_graph_strategy compilation setting and automatic RTX-native CUDA graph integration for the Python runtime path (PythonTorchTensorRTModule).

TensorRT-RTX has native CUDA graph support via IRuntimeConfig.cuda_graph_strategy, where the JIT compiler handles capture/replay/invalidation internally. This is superior to manual torch.cuda.CUDAGraph() capture on RTX because:

• Manual capture freezes fallback kernels; lazy-compiled specialized kernels can never replace them
• Runtime allocation or data-dependent shapes can cause cudaStreamBeginCapture to fail
• The JIT compiler automatically manages graph staleness (shape changes, pointer changes, kernel readiness)

Key changes

• New cuda_graph_strategy setting on CompilationSettings ("disabled" / "whole_graph_capture")
• Mapped to trt.CudaGraphStrategy on IRuntimeConfig (same pattern as dynamic_shapes_kernel_specialization_strategy)
• SUBGRAPH mode (set_cudagraphs_mode(True)): On RTX, always use RTX-native CUDA graphs — manual capture is bypassed. If cuda_graph_strategy was not explicitly set, the runtime overrides to whole_graph_capture and warns.
• WHOLE_GRAPH mode (enable_cudagraphs() with mixed TRT + PyTorch ops): Validates all TRT engines are monolithically capturable via context.is_stream_capturable(stream) and strategy != "lazy". If capturable, proceeds with outer monolithic capture (RTX-native disabled per-engine). If not capturable, raises RuntimeError.
• _is_monolithic_capturable() — runtime check combining stream capturability and kernel specialization strategy
• _enable_rtx_native_cudagraphs() — recreates execution context with WHOLE_GRAPH_CAPTURE
• _check_monolithic_capturability() in CudaGraphsTorchTensorRTModule for mixed graph validation

Behavior matrix

Graph type	cudagraph mode	RTX?	Behavior
TRT-only	SUBGRAPH	Yes	RTX-native always (override if needed)
TRT-only	SUBGRAPH	No	Manual capture (existing)
Mixed	WHOLE_GRAPH	Yes + capturable	Monolithic capture; RTX-native disabled per-engine
Mixed	WHOLE_GRAPH	Yes + NOT capturable	RuntimeError
Mixed	WHOLE_GRAPH	No	Monolithic capture (existing)
Any	No cudagraphs + strategy set	Yes	RTX-native runs transparently

Depends on #4180 (runtime cache) and #4184 (dynamic shapes strategy).

Type of change

• New feature (non-breaking change which adds functionality)

Checklist:

• My code follows the style guidelines of this project (You can use the linters)
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas and hacks
• I have made corresponding changes to the documentation
• I have added tests to verify my fix or my feature
• New and existing unit tests pass locally with my changes
• I have added the relevant labels to my PR in so that relevant reviewers are notified