feat(BEVFusion): export bevfusion-camera onnx

[KSeangTan]

Summary

This PR updates the deployment script in BEVFusion for LiDAR and Camera-only. This PR also cleans up the code in torch2onnx to make it more modular.

Change point

• Fix the bug when running lidar-only deployment and points is not available [1]
• Add projects/BEVFusion/configs/deploy/bevfusion_camera_point_bev_tensorrt_dynamic.py for bevfusion_camera_only deployment config [2]
• Add projects/BEVFusion/configs/deploy/bevfusion_main_body_lidar_only_intensity_tensorrt_dynamic.py for bevfusion_lidar_only with intensities [3]
• Add projects/BEVFusion/deploy/builder.py to move all code related to ONNX deployment configs to this file [4]
• Add TrtBevFusionCameraOnlyContainer to container for the camera-only network [5]
• Add projects/BEVFusion/deploy/exporter.py to move all code related to model export to this file [5]
• Add projects/BEVFusion/deploy/utils.py to move all code related to config setup to this file [6]
• Update VoxelDetection to include the cases for images and lidar points [7], [8], [9]

Test performed

• Both camera-only and lidar-only successfully generates onnx files, and lidar-only onnx was tested in Autoware. However, the camera-only is not tested yet since we need to make few changes first.

[amadeuszsz]

Overall LGTM, minor comments.
Please consider if we can update README file 🙏🏻.

[KSeangTan]

Overall LGTM, minor comments. Please consider if we can update README file 🙏🏻.

Thanks! It's updated here: 78f4337