feat: random crops for object detector training with torch backend

fix(tensorrt): yolox postprocessing in C++

Onnx post processing returns wrong results due to a tensorrt bug

fix: CI on Jetson nano with lighter classification model

fix(trt): apply std to input image correctly

feat(ml): data augmentation for object detection with torch backend

fix(build): docker builds with tcmalloc

feat(torch): SWA for RANGER/torch (https://arxiv.org/abs/1803.05407)

fix: tag syntax for github releases

chore(release): 0.13.0

chore(release): 0.12.0