Re-organize online inferene, support more models, add demo results, update docs, add test

[SamitHuang]

…pdate docs, add test

Thank you for your contribution to the MindOCR repo.
Before submitting this PR, please make sure:

• You have read the Contributing Guidelines on pull requests
• Your code builds clean without any errors or warnings
• You are using approved terminology
• You have added unit tests

Motivation

1. Add scripts to support detection inference, rec inference, and system inference individually
2. Allow inference on a single image, with no limit to directory input, like python predict_system.py --image_dir test.jpg
3. Align command line APIs and arguments to paddleocr inference for low-coat transfer.
4. Remove redundant font file. font files are all placed in docs/font/xx.ttf
5. Previous detection preprocessing only fit the icdar15 dataset (that's input shape is 720x1280), but can not work correctly for input images with arbitrary shapes. This code adopt the new DetResize to fix this problem, allowing setting limit_type as min or max to better resize for inference.
6. Add some visualization results on EN and CH text images, which can make the doc more vivid.

Test Plan

Add system test for it.

Already test the new pipeline on icdar 15 test set. Results are shown in tools/infer/text/README.md, which is slightly better than previous version with the same det and rec algorithm. The running time is also similar, which is about 94s for inference on the whole icdar15 detection test set.

Related Issues and PRs

(Is this PR part of a group of changes? Link the other relevant PRs and Issues here. Use https://help.github.com/en/articles/closing-issues-using-keywords for help on GitHub syntax)