Dynamo export and improve benchmark script for SAM2 encoder (#23887)

### Description
* Add dynamo export for Sam2 image encoder
* Verify fp32 onnx model with CPU EP (to avoid error message from TRT
EP).
* Update benchmark script:
  - output ORT profiling
- output torch compiled code and unique kernel name for compiled kernel
  - add an option for nightly package installation
  - uninstall existing ort packages before installing

The node metadata of dynamo exported model can help mapping node in onnx
model back to pytorch modeling script. Currently, the graph optimization
is not done on dynamo exported model, so it is experimental right now.

### Motivation and Context

To support profiling of torch compiled CUDA kernel.

Author: tianleiwu

onnxruntime/python/tools/transformers/models/sam2/README.md

@@ -96,8 +96,7 @@ We can create a conda environment then run GPU benchmark like the following:
 conda create -n sam2_gpu python=3.11 -y
 conda activate sam2_gpu
 install_dir=$HOME
-profiling=true
-bash benchmark_sam2.sh $install_dir gpu $profiling
+bash benchmark_sam2.sh $install_dir gpu
 ```
 
 or create a new conda environment for CPU benchmark:
@@ -107,16 +106,28 @@ conda activate sam2_cpu
 bash benchmark_sam2.sh $HOME cpu
 ```
 
-The first parameter is a directory to clone git repositories or install CUDA/cuDNN for benchmark.
-The second parameter can be either "gpu" or "cpu", which indicates the device to run benchmark.
-The third parameter is optional. Value "true" will enable profiling after running benchmarking on GPU.
+The usage of the script like the following:
+```
+bash benchmark_sam2.sh <install_dir> <cpu_or_gpu> [profiling] [benchmarking] [nightly] [dynamo]
+```
+
+| Parameter| Default  | Description |
+|----------|----------| ------------|
+| install_dir | $HOME | a directory to clone git repositories or install CUDA/cuDNN for benchmark |
+| cpu_or_gpu | gpu | the device to run benchmark. The value can be either "gpu" or "cpu" |
+| profiling | false | run gpu profiling |
+| benchmarking | true | run benchmark |
+| nightly | false | install onnxruntime nightly or official release package |
+| dynamo | false | export image encoder using dynamo or not. |
 
-The script will automatically install required packages in current conda environment, download checkpoints, export onnx,
-and run demo, benchmark and optionally run profiling.
+The dynamo export is experimental since graph optimization still need extra works for this model.
 
-* The performance test result is in sam2_gpu.csv or sam2_cpu.csv, which can be loaded into Excel.
-* The demo output is sam2_demo_fp16_gpu.png or sam2_demo_fp32_cpu.png.
-* The profiling results are in *.nsys-rep or *.json files in current directory. Use Nvidia NSight System to view the *.nsys-rep file.
+Output files:
+* sam2_cpu_[timestamp].csv or sam2_gpu_[timestamp].csv has benchmark results. Use Excel to load the file to view it.
+* onnxruntime_image_[encoder|decoder].json has ONNX Runtime profiling results. Use `chrome://tracing` in Chrome browser to view it.
+* torch_image_[encoder|decoder].json has PyTorch profiling results. Use `chrome://tracing` in Chrome browser to view it.
+* sam2_fp16_profile_image_[encoder|decoder]_[ort|torch]_gpu.[nsys-rep|sqlite] has NVTX profiling. Use Nvidia NSight System to view it.
+* torch_image_encoder_compiled_code.txt has the compiled kernel code from Pytorch.
 
 ## Limitations
 - The exported image_decoder model does not support batch mode for now.

onnxruntime/python/tools/transformers/models/sam2/benchmark_sam2.py

@@ -46,6 +46,7 @@ def __init__(
         prefer_nhwc: bool = False,
         warm_up: int = 5,
         enable_nvtx_profile: bool = False,
+        enable_ort_profile: bool = False,
         enable_torch_profile: bool = False,
         repeats: int = 1000,
         verbose: bool = False,
@@ -74,6 +75,7 @@ def __init__(
         self.prefer_nhwc = prefer_nhwc
         self.warm_up = warm_up
         self.enable_nvtx_profile = enable_nvtx_profile
+        self.enable_ort_profile = enable_ort_profile
         self.enable_torch_profile = enable_torch_profile
         self.repeats = repeats
         self.verbose = verbose
@@ -317,6 +319,7 @@ def run_test(
         repeats=args.repeats,
         warm_up=args.warm_up,
         enable_nvtx_profile=args.enable_nvtx_profile,
+        enable_ort_profile=args.enable_ort_profile,
         enable_torch_profile=args.enable_torch_profile,
         torch_compile_mode=args.torch_compile_mode,
         verbose=False,
@@ -325,7 +328,7 @@ def run_test(
     if args.engine == "ort":
         sess_options = SessionOptions()
         sess_options.intra_op_num_threads = args.intra_op_num_threads
-        if config.enable_nvtx_profile:
+        if config.enable_ort_profile:
             sess_options.enable_profiling = True
             sess_options.log_severity_level = 4
             sess_options.log_verbosity_level = 0
@@ -349,6 +352,8 @@ def run_test(
             with nvtx.annotate("one_run"):
                 _ = session.infer(input_dict)
             cudart.cudaProfilerStop()
+
+        if config.enable_ort_profile:
             session.ort_session.end_profiling()
 
         if repeats == 0:
@@ -554,6 +559,14 @@ def _parse_arguments():
         help="Enable nvtx profiling. It will add an extra run for profiling before performance test.",
     )
 
+    parser.add_argument(
+        "--enable_ort_profile",
+        required=False,
+        default=False,
+        action="store_true",
+        help="Enable ORT profiling.",
+    )
+
     parser.add_argument(
         "--enable_torch_profile",
         required=False,