efficientNMSPlugin: support class-independent nms with new parameter class_agnostic

[yili123123]

@rajeevsrao @pranavm-nvidia @samurdhikaru ,sorry for closing the previous pull request by mistake. Could you take your time to review the new PR?

Some object detection networks/architectures like YOLO series need to use class independent nms operations. Now we can do class independent nms when the class_agnostic is enabled, otherwise, different class would do nms respectively.

[tadejsv]

Would really love to see this merged! This functionality would bring the behavior of NMS plugin in line with ONNX and pytorch.

[rajeevsrao]

@yili123123 thanks for your contribution! We will merge this once the internal review/testing (currently in progress) is completed. @samurdhikaru will coordinate.

[samurdhikaru]

@yili123123 Please refer to 76eaca2 to see whether it fulfils the intended purpose. If not, please update the PR accordingly.

[yili123123]

@yili123123 Please refer to 76eaca2 to see whether it fulfils the intended purpose. If not, please update the PR accordingly.

@samurdhikaru Okay, it fully meets expectations, is there anything else I need to do next?