[FEATURE]Horovod support for training transformer + add mirror data for wmt (PART 1)

[hutao965]

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

• horovod support for training transformer and add split option for BoundedBugdetSampler
• add mirror wmt data source

[codecov]

Codecov Report

Merging #1284 into numpy will increase coverage by 0.08%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##            numpy    #1284      +/-   ##
==========================================
+ Coverage   83.60%   83.68%   +0.08%     
==========================================
  Files          42       42              
  Lines        6375     6382       +7     
==========================================
+ Hits         5330     5341      +11     
+ Misses       1045     1041       -4     

Impacted Files	Coverage Δ
src/gluonnlp/data/sampler.py	96.22% <100.00%> (+0.10%)	⬆️
src/gluonnlp/layers.py	86.78% <0.00%> (ø)
src/gluonnlp/embedding/embed_loader.py	81.52% <0.00%> (ø)
src/gluonnlp/utils/misc.py	44.69% <0.00%> (+1.28%)	⬆️

[sxjscience]

@hymzoque The horovod PR has been merged. Would you merge master?

[sxjscience]

@szhengac Does the Transformer-Big results look reasonable to you?

[sxjscience]

LGTM in general, you may add some transformer_base numbers.

[sxjscience]

I'll approve and merge for now. Still need to wait for PART2, which adds horovod command and results of TransformerBase.