Introduce faster tokenizer for BERT

[cmdevries]

Description

This change introduces a new tokenizer for BERT that is 3.5x faster on a
2017 13 inch MacBook pro.

It was tested by tokenizing the test string u"UNwant\u00E9d,running"
from test_transforms.py::bert_tokenizer 100,000 times using the timeit
module.

The existing implementation with the Cython optmized wordpiece took
5.56 seconds and the new implementation took 1.58 seconds.

The changes were originally authored by Eric Lind [email protected]
and this commit integrates them with Gluon NLP.

Checklist

Essentials

• PR's title starts with a category (e.g. [BUGFIX], [MODEL], [TUTORIAL], [FEATURE], [DOC], etc)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage
• Code is well-documented

Changes

• Feature1, tests, (and when applicable, API doc)
• Feature2, tests, (and when applicable, API doc)

Comments

• If this change is a backward incompatible change, why must this change be made.
• Interesting edge cases to note here

cc @dmlc/gluon-nlp-team

[codecov]

Codecov Report

Merging #1024 into master will increase coverage by 20.54%.
The diff coverage is 100%.

@@             Coverage Diff             @@
##           master    #1024       +/-   ##
===========================================
+ Coverage   67.74%   88.29%   +20.54%     
===========================================
  Files          67       67               
  Lines        6340     6261       -79     
===========================================
+ Hits         4295     5528     +1233     
+ Misses       2045      733     -1312

Impacted Files	Coverage Δ
src/gluonnlp/data/transforms.py	82.58% <100%> (+60.28%)	⬆️
src/gluonnlp/data/word_embedding_evaluation.py	96.94% <0%> (+0.76%)	⬆️
src/gluonnlp/model/train/language_model.py	96.21% <0%> (+1.62%)	⬆️
src/gluonnlp/model/sequence_sampler.py	35.88% <0%> (+3.13%)	⬆️
src/gluonnlp/embedding/evaluation.py	95.79% <0%> (+4.2%)	⬆️
src/gluonnlp/model/transformer.py	91.31% <0%> (+4.5%)	⬆️
src/gluonnlp/model/bilm_encoder.py	100% <0%> (+5.08%)	⬆️
src/gluonnlp/model/convolutional_encoder.py	97.67% <0%> (+6.97%)	⬆️
src/gluonnlp/model/parameter.py	100% <0%> (+8%)	⬆️
src/gluonnlp/data/stream.py	84.97% <0%> (+8.29%)	⬆️
... and 33 more

[mli]

Job PR-1024/6 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-1024/6/index.html

[mli]

Job PR-1024/7 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-1024/7/index.html

[mli]

Job PR-1024/8 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-1024/8/index.html

[leezu]

LGTM, thank you!