Optimizing the output of classification probablity in demo\text_classification\predict.py and corresponding README.md

Mismatching between result prediction and the corresponding code in demo\sequence_labeling\README.md

* update Transformer module that can do text-matching task

* Add dataset, module task, and demo of text-matching

* Add embedding seq-cls finetune demo and update api

* Update docs of pad_sequence and trunc_sequence

* Add electra modules

* Amend module versions of chinese-bert-wwm, chinese-bert-wwm-ext, rbt3 and rbtl3

* Updata demo README.md

* Add tutorial about nlp dataset customization

* Update doc and tokenizer usage in seq-cls demo

* Updata tutorial

* Add token-cls task for transformer modules

* Fix numpy dtype mismatch in windows

* Update README.md

* Update token-cls task for ernie_tiny

* Update token-cls task for ernie_tiny

* Update token-cls task for other Transformer modules

* Update README.md of modules and token-cls demo

* Add chunk_scheme arg check in SeqLabelingDataset

* Update ChunkEvaluator and paddlenlp requirement

* Update README.md

* Update token-cls demo

* Add MSRA dataset

* Add ernie_tiny seq-cls task case

* Update README.md