[FEATURE] Support YOLOv6 3.0 inference

[Qingrenn]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

Support YOLOv6 3.0 inference.

Modification

• add CSPSPPFBottleneck module for YOLOv6 3.0 backbone

• add BiFusion module, YOLOv6RepBiPAFPN module for YOLOv6 3.0 neck

• modify YOLOv6HeadModule to support YOLOv6 3.0 head

• add a yolov6v3 l/m/s/t/n conifgs

• add a checkpoint convert script

BC-breaking (Optional)

Use cases (Optional)

1. Transform the official checkpoints to mmyolo style.
2. perform tool/test.py to evaluate the models.

The results on COCO val2017 using the transformed checkpoints:

Model	box_mAP	box_mAP_50	bbox_mAP_75	bbox_mAP_s	bbox_mAP_m	bbox_mAP_l
yolov6_v3_n	0.3750	0.5300	0.4060	0.1780	0.4180	0.5490
yolov6_v3_s	0.4520	0.6210	0.4870	0.2490	0.5040	0.6260
yolov6_v3_m	0.5000	0.6710	0.5440	0.3080	0.5540	0.6670
yolov6_v3_l	0.5290	0.7050	0.5780	0.3470	0.5820	0.6990

Checklist

1. Pre-commit or other linting tools are used to fix potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit tests to ensure the correctness.
3. If the modification has a potential influence on downstream projects, this PR should be tested with downstream projects, like MMDetection or MMClassification.
4. The documentation has been modified accordingly, like docstring or example tutorials.

ToDo👨🏻‍💻

update loss of v3.0 for training.

[CLAassistant]

All committers have signed the CLA.

[RangeKing]

@Qingrenn Thank you for your contribution.

• Please sign the CLA by clicking this button.
  
• The config name should be like yolov6_s-v3_syncbn_fast_8xb32-400e_coco.py.