Seqal

[PyTorch v1.9.1](https://pypi.org/project/torch/1.9.1/#files) doesn't provide arm64 (Mac M1) wheel, so we update PyTorch to [v1.10.0](https://pypi.org/project/torch/1.10.0/#files).

We refactor samplers to make them easy to use and update [Tutorial 3: Active Learner Setup](https://github.com/tech-sketch/SeqAL/blob/v0.3.0/docs/TUTORIAL_3_Active_Learner_Setup.md)

Changes:
- update packages
- update readme and tutorial
- add Aligner, Transformer, PerformanceRecorder
- refactor ActiveLearner, Stopper, etc.

Tutorials give more detail about this update.

- [Tutorial 1: Introduction](docs/TUTORIAL_1_Introduction.md)
- [Tutorial 2: Prepare Corpus](docs/TUTORIAL_2_Prepare_Corpus.md)
- [Tutorial 3: Active Learner Setup](docs/TUTORIAL_3_Active_Learner_Setup.md)
- [Tutorial 4: Prepare Data Pool](docs/TUTORIAL_4_Prepare_Data_Pool.md)
- [Tutorial 5: Research and Annotation Mode](docs/TUTORIAL_5_Research_and_Annotation_Mode.md)
- [Tutorial 6: Query Setup](docs/TUTORIAL_6_Query_Setup.md)
- [Tutorial 7: Annotated Data](docs/TUTORIAL_7_Annotated_Data.md)
- [Tutorial 8: Stopper](docs/TUTORIAL_8_Stopper.md)
- [Tutorial 9: Ouput Labeled Data](docs/TUTORIAL_9_Ouput_Labeled_Data.md)
- [Tutorial 10: Performance Recorder](docs/TUTORIAL_10_Performance_Recorder.md)
- [Tutorial 11: Multiple Language Support](docs/TUTORIAL_11_Multiple_Language_Support.md)

Move the sequence tagger inside active learner and add the token-based conter.

With v2.0.0, you can use seqal to build your own active learning workflow for sequence labeling task.