[TensorRT EP] Support engine hardware compatibility

[yf711]

Description

• Introduce option trt_engine_hw_compatible to support engine hardware compatibility for Ampere+ GPUs
  • This enables nvinfer1::HardwareCompatibilityLevel::kAMPERE_PLUS flag when generating engines
  • This option has been validated on sm80/86 GPUs, as engine can be reused across different ampere+ arch:
    • Client side need to enable this option as well to leverage existing sm80+ engines
  • If this option is enabled by users which TRT<8.6 or sm<80, there will be a warning showing this option not supported

Engine naming:

When	trt_engine_hw_compat=false	trt_engine_hw_compat=true
A100 (sm80)	TensorrtExecutionProvider_TRTKernel_graph_torch-jit-export_9454133937466702238_0_0_sm80.engine	TensorrtExecutionProvider_TRTKernel_graph_torch-jit-export_9454133937466702238_0_0_sm80+.engine
RTX3080 (sm86)	TensorrtExecutionProvider_TRTKernel_graph_torch-jit-export_9454133937466702238_0_0_sm86.engine	TensorrtExecutionProvider_TRTKernel_graph_torch-jit-export_9454133937466702238_0_0_sm80+.engine

Motivation and Context

Reference: https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#hardware-compat

[chilo-ms]

A minor issue here:
In the case of Embedded engine model / EPContext model, the TensorRTCacheModelHandler::ValidateEPCtxNode() will check the compute capability between the underlying GPU and "hardware_architecture" attribute in the EPContext node, and throw warning message if they are not matched.
Now TRT EP supports Ampere+ hardware compatibility, we might need to modify the check function there if the embedded engine cache name has "sm80+".