[PRX] Improving export to Tensorrt

[WaterKnight1998]

Is your feature request related to a problem? Please describe.

When I export PRX model to ONNX and try to build TensorRT Engine it fails because Col2Im operator is not supported. This Col2Im is common when using torch.nn.functional.fold and torch.nn.functional.unfold. Those functions are used in img2seq function and in seq2img

Describe the solution you'd like.
Reimplement those functions so export to onnx generates graph without thos operations.