[Model] PGNet Model Support

[wxsIcey]

1. Background
   MindOCR currently does not support end-to-end detection and recognition models, I hope to solve this problem by contributing a PGNet model.

2. Model Introduction
   Existing text recognizers are mostly built on two-stage frameworks or character-based methods, which suffer from non-maximum suppression (NMS), region of interest (RoI) operations or character-level annotations. To solve the above problems, PGNets uses a novel fully convolutional point collection network for reading arbitrary-shaped text in real time.

3. Reference Papers
   PGNet: Real-time Arbitrarily-Shaped Text Spotting with Point Gathering Network Pengfei Wang, Chengquan Zhang, FeiQi, Shanshan Liu, Xiaoqiang Zhang, Pengyuan Lyu, Junyu Han, Jingtuo Liu, Errui Ding, Guangming Shi AAAI, 2021

4. Expectations
   I'd like to get the community's comments, suggestions, and related reference documents or PRs.