[BUGFIX] Corrected np.random.randint upper limit in data.stream.py

[ggbaro]

Description

Since default np.int_ is int32 for Windows, an upper limit for random seed of 2**32 rises a ValueError: high is out of bounds for int32.
with np.iinfo(np.int_).max, np.random.randint is platform-independent

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

• very small fix to allow Windows users to use module (otherwise random seed is bugged)

cc @dmlc/gluon-nlp-team

[codecov]

Codecov Report

Merging #935 into master will increase coverage by 1.2%.
The diff coverage is 100%.

@@            Coverage Diff            @@
##           master     #935     +/-   ##
=========================================
+ Coverage   89.51%   90.71%   +1.2%     
=========================================
  Files          67       67             
  Lines        6360     6389     +29     
=========================================
+ Hits         5693     5796    +103     
+ Misses        667      593     -74

Impacted Files	Coverage Δ
src/gluonnlp/data/stream.py	84.97% <100%> (ø)	⬆️
src/gluonnlp/model/train/language_model.py	95.33% <0%> (-1.83%)	⬇️
src/gluonnlp/model/language_model.py	98.54% <0%> (-1.46%)	⬇️
src/gluonnlp/model/train/cache.py	97.82% <0%> (+0.2%)	⬆️
src/gluonnlp/model/attention_cell.py	94.55% <0%> (+1.36%)	⬆️
src/gluonnlp/model/transformer.py	91.2% <0%> (+1.5%)	⬆️
src/gluonnlp/model/block.py	53.19% <0%> (+2.12%)	⬆️
src/gluonnlp/data/utils.py	76.33% <0%> (+2.29%)	⬆️
src/gluonnlp/model/sequence_sampler.py	91.63% <0%> (+17.07%)	⬆️
src/gluonnlp/model/seq2seq_encoder_decoder.py	75% <0%> (+29.68%)	⬆️

[mli]

Job PR-935/1 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-935/1/index.html

[szha]

It seems that the test_dataloader test is consistently hanging in the CI. For example: http://ci.mxnet.io/blue/organizations/jenkins/GluonNLP-py3-cpu-unittest/detail/PR-935/3/pipeline#step-159-log-1116

[leezu]

Could be due to using 64 bit instead of 32 bit range used before. I added a commit so that np.int32 is used

[leezu]

np.int_ can be 64 bit, and it is 64 bit on our CI

[mli]

Job PR-935/2 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-935/2/index.html