YOLOv8/11 postprocess support exp

[ret7020]

Yolov8/11 models have new head architecture, so current postproc implementation in coco_mAP.py don't work. I created experimental postproc_yolov8 function that can handle yolov11 model (I don't check it yet on multi-class models and yolov8, but I will check it later).

I modified model_eval.py --postprocess_type arg to select default postproc if it coco_mAP and use my implementation if arg is coco_mAP_yolov8.

If this approach will be approved I will tune other code (because I have problem with using visual_tool, I think problem also in old post processing).

And also I use following model export from pt to onnx:

from ultralytics import YOLO
import types
import sys

input_size = (int(sys.argv[2]), int(sys.argv[3]))

def forward2(self, x):
  x_reg = [self.cv2[i](x[i]) for i in range(self.nl)]
  x_cls = [self.cv3[i](x[i]) for i in range(self.nl)]
  return x_reg + x_cls

model_path = sys.argv[1]
model = YOLO(model_path)
model.model.model[-1].forward = types.MethodType(forward2, model.model.model[-1])
model.export(format='onnx', opset=11, imgsz=input_size)

This custom head forward method used in CV181x docs (and TDL SDK implementation for CV181x also waits for such outputs)