Add RARE (CRNN-seq2seq) README and support O2 training

[zhtmike]

• Add RARE (CRNN-seq2seq) README and benchmark result.
• The native GRUCell operator does not support FP16 input (see issue https://gitee.com/mindspore/mindspore/issues/I6ZY2O), here we implemented a input-fixed version
• Update the multi dataset evaluation script due to the interface update (num_columns -> columns_index), handle the exception when data_dir_root is incorrect

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

(Write your motivation for proposed changes here.)

Test Plan

(How should this PR be tested? Do you require special setup to run the test or repro the fixed bug?)

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)

[SamitHuang]

1. 可否通过修改amp.py里的white/black list来将RNN层设成半精度？
2. 建议rename _nn -> utils，可把自定义的layers 和 helpers放里面

[zhtmike]

RNNCell应当属于white list, 会把input 自动cast为fp16。但目前mindspore RNNCell会在fp16输入时前向报错，原因是RNNCell内参如weight_ih没有自动转换成fp16，从而当用到MatMul算子时因为两边输入type不对齐而报错。此改动强制将内参和输入转换成统一type，避免这一问题。

[SamitHuang]

了解，那应该在自定义的RNNCell init中加入 cast_fp16的选项？否则如果设成O0的话，该层仍然会半精度计算。

[zhtmike]

现在RNNCell内参会根据输入的type改变的，如果输入为fp32，那内参会改为fp32，计算会是全精度

[zhtmike]

_nn 已才成 utils

[SamitHuang]

1. 可否通过修改amp.py里的white/black list来将RNN层设成半精度？
2. 建议rename _nn -> utils，可把自定义的layers 和 helpers放里面

[zhtmike]

Uploaded checkpoint path, can review @SamitHuang @HaoyangLee

[SamitHuang]

Conflicts are to be solved before merge