Add comprehensive logging for TensorRT compilation process

[alexnorell]

Summary

• Adds detailed logging throughout the TensorRT compilation pipeline to provide better visibility into the compilation process
• Tracks and reports compilation timing (which can take several minutes)
• Logs runtime environment, configuration details, and progress for each phase

Changes

engine_builder.py

• Added ONNX parsing progress logs (start, parsing, completion)
• Enhanced engine compilation logging with:
  • Visual separators for clear section boundaries
  • Configuration details (precision, input size, batch settings, compatibility flags)
  • Platform capability checks for FP16/INT8 support
  • Optimization profile configuration status
  • Clear "Building TensorRT engine - this may take several minutes..." message before the slow compilation step
  • Build time tracking with elapsed seconds
  • Engine size in MB
  • Completion confirmation

core.py

• Added high-level compilation orchestration logging:
  • Compilation start message with model directory
  • Runtime environment capture notification
  • GPU availability and device information (CUDA version, TensorRT version)
  • ONNX model loading notification
  • Detection and logging of pre-existing engines (compilation skip)
  • TRT configuration save location
  • Engine builder initialization with workspace size
  • Total compilation time in both seconds and minutes

Example Output

Starting TRT compilation for model in: /path/to/model
Capturing runtime environment information...
GPU Available: True
GPU Devices: ['NVIDIA GeForce RTX 3090']
CUDA Version: 11.8
TensorRT Version: 8.6.1
Loading ONNX model from: /path/to/model/weights.onnx
Starting ONNX parsing from: /path/to/model/weights.onnx
Parsing ONNX model graph...
ONNX parsing completed successfully
============================================================
Starting TensorRT Engine Compilation
============================================================
Output path: /path/to/model/engine-fp16.plan
Precision: FP16
Input size: 640x640
Using static batch size
Configuring builder flags...
FP16 is supported on this platform
Creating optimization profile...
Building TensorRT engine - this may take several minutes...
TensorRT engine built successfully in 123.45 seconds
Engine size: 45.23 MB
============================================================
TensorRT Compilation Complete
============================================================
Total compilation time: 125.67 seconds (2.09 minutes)

Test plan

• Verify logs appear during TensorRT compilation
• Confirm timing information is accurate
• Check that all compilation phases are logged
• Validate that existing engines are detected and skip compilation is logged