Remove NLTKMosesTokenizer in favor of SacreMosesTokenizer

[leezu]

Description

NLTKMosesTokenizer works only with unsupported / outdated nltk==3.2.5 but was
kept in GluonNLP as SacreMosesTokenizer does not support Python 2. Drop it as
Python 2 support has been dropped.

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

• Remove NLTKMosesTokenizer in favor of SacreMosesTokenizer
• Remove NLTKMosesDetokenizer in favor of SacreMosesDetokenizer

Comments

• By design this is backwards incompatible.

cc @dmlc/gluon-nlp-team

[mli]

Job PR-942/2 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-942/2/index.html

[mli]

Job PR-942/1 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-942/1/index.html

[codecov]

Codecov Report

Merging #942 into master will increase coverage by 0.47%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #942      +/-   ##
==========================================
+ Coverage   90.12%   90.59%   +0.47%     
==========================================
  Files          67       67              
  Lines        6389     6380       -9     
==========================================
+ Hits         5758     5780      +22     
+ Misses        631      600      -31

Impacted Files	Coverage Δ
src/gluonnlp/data/transforms.py	86% <100%> (+9%)	⬆️
src/gluonnlp/model/bert.py	84.95% <0%> (-14.51%)	⬇️
src/gluonnlp/data/word_embedding_evaluation.py	96.96% <0%> (+0.75%)	⬆️
src/gluonnlp/data/dataset.py	99.2% <0%> (+1.58%)	⬆️
src/gluonnlp/data/batchify/batchify.py	96.59% <0%> (+3.4%)	⬆️
src/gluonnlp/data/corpora/wikitext.py	100% <0%> (+5.17%)	⬆️
src/gluonnlp/model/parameter.py	100% <0%> (+8%)	⬆️

[mli]

Job PR-942/3 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-942/3/index.html

[sxjscience]

Let's raise a warning instead of removing the code.

[leezu]

People that use the NLTKMosesTokenizer or NLTKMosesDetokenizer class may break LGPL without noticing. It can have severe impact, thus I think it's better to remove the code. For example, in a corporate setting usage of LGPL software typically requires special approval.