Support skipping preprocess when exporting ONNX

[zhiqwang]

Using the following snippet will export a dynamic batch/shape ONNX model without pre-processing (aka letterbox in YOLOv5 and NestedTensor in yolort).

# 'yolov5s.pt' is downloaded from https://github.com/ultralytics/yolov5/releases/download/v5.0/yolov5s.pt
python tools/export_model.py --checkpoint_path yolov5s.pt --skip_preprocess

https://github.com/zhiqwang/yolov5-rt-stack/blob/6cb22845a41bf4dd846a9bcad06c09d3fec8cd49/yolort/models/transform.py#L217-L251

Meanwhile we require the shape of the inputs to be Nx3xHxW, and as such this will close #39, close #45 and close #159.

[codecov]

Codecov Report

Merging #193 (5421eea) into master (6cb2284) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #193   +/-   ##
=======================================
  Coverage   96.48%   96.49%           
=======================================
  Files          10       10           
  Lines         598      599    +1     
=======================================
+ Hits          577      578    +1     
  Misses         21       21           

Flag	Coverage Δ
unittests	96.49% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
test/test_utils.py	98.41% <100.00%> (+0.02%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6cb2284...5421eea. Read the comment docs.