Add profiling features and onnx cleaner

Context

Related to aidge_onnx#83 (moved). Enhances TensorRT export module with comparative analysis of raw and cleaned models using ONNX cleaner and other frameworks.

Current limitations: external modules (aidge_onnx/onnxsim/onnxslim) should be installed in the running container for now.

Current changes

• Profiling API: Added getProfileData() and getAverageExecutionTime() methods to Graph class
• Dedicated benchmark script: New scripts/benchmark.py containing the benchmark function for better separation of concerns :
  • Flexible benchmarking: with/without ONNX cleaners (onnx_cleaner, onnxsim, onnxslim)
  • Performance metrics: execution time, throughput, speedup analysis
• Simple test script: test.py to execute profiling and optionally benchmark cleaners

TODO

• Profiling API
• Basic compare script
• Integration strategy decision
• Clean docker installation (Manual installation required)
• Documentation