[FEATURE] update backtranslation and add multinomial sampler

[hutao965]

Description

update some scripts to support backtranslation better and add a multinomial sampler

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

• supports cleaning mono corpus
• supports --max_update in train_transformer.py
• provides backtranslation script
• supports multinomial sampler in sequence_sampler.py
• passed test case of multinomial sampler in test_sequence_sampler.py

cc @dmlc/gluon-nlp-team

[codecov]

Codecov Report

Merging #1259 into numpy will increase coverage by 0.23%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##            numpy    #1259      +/-   ##
==========================================
+ Coverage   82.32%   82.56%   +0.23%     
==========================================
  Files          38       38              
  Lines        5410     5488      +78     
==========================================
+ Hits         4454     4531      +77     
- Misses        956      957       +1     

Impacted Files	Coverage Δ
src/gluonnlp/sequence_sampler.py	86.32% <100.00%> (+2.65%)	⬆️
src/gluonnlp/utils/misc.py	48.18% <0.00%> (-2.61%)	⬇️
src/gluonnlp/data/filtering.py	78.26% <0.00%> (ø)
src/gluonnlp/utils/preprocessing.py	100.00% <0.00%> (ø)
src/gluonnlp/utils/lazy_imports.py	55.71% <0.00%> (+1.02%)	⬆️
src/gluonnlp/data/loading.py	83.39% <0.00%> (+7.54%)	⬆️

[sxjscience]

Also need to remove the TODO item here:

gluon-nlp/src/gluonnlp/sequence_sampler.py

Lines 69 to 73 in cd48efd

	# TODO(sxjscience)
	# 1. Add Multinomial Sampler with Temperature
	# 2. Add Nucleus Sampling "[ICLR2020] The Curious Case of Neural Text Degeneration"
	# (https://openreview.net/pdf?id=rygGQyrFvH)
	# 3. Add ParticleFilter Sampler

and add the paper reference to top-p sampling.