Upgrade to PyTorch 0.4 + Small Fixes

[PetrochukM]

This PR adds PyTorch 0.4 support.

Updates:

• Removed Variable were appropriate
• Support for Tensor scalars and item() API
• Use context managers instead of volatile
• Updated .new to the tensor.new_* API
• Updated .travis build scripts to test for PyTorch 0.4
• Updated clip_grad_norm to clip_grad_norm_
• Updated constant to constant_
• Updated Variable(tensor.data) pattern to .detach()

Other:

• Removed embedded_dropout from the examples. It relied on old PyTorch 0.3.1 private API features and was not relevant to using PyTorch-NLP.
• Added a 1% threshold to codecov allowing a small decrease in code coverage.
• Removed now that multi30k @pytest.mark.skip(reason="Unsafe dataset host (SSL: CERTIFICATE_VERIFY_FAILED)")
• Added test to LockDrop
• Removed random prints
• Updated to sacremoses to fix NLTK moses dependancy

[codecov-io]

Codecov Report

Merging #29 into master will increase coverage by 0.03%. The diff coverage is 88.23%.

@@            Coverage Diff             @@
##           master      #29      +/-   ##
==========================================
+ Coverage   94.59%   94.63%   +0.03%     
==========================================
  Files          55       55              
  Lines        1536     1528       -8     
==========================================
- Hits         1453     1446       -7     
+ Misses         83       82       -1

Impacted Files	Coverage Δ
torchnlp/nn/attention.py	96.66% <ø> (ø)	:arrow_up:
torchnlp/utils.py	93.06% <ø> (ø)	:arrow_up:
torchnlp/word_to_vector/pretrained_word_vectors.py	78.82% <ø> (-0.25%)	:arrow_down:
torchnlp/nn/weight_drop.py	100% <100%> (ø)	:arrow_up:
torchnlp/metrics/accuracy.py	100% <100%> (ø)	:arrow_up:
torchnlp/nn/lock_dropout.py	93.33% <100%> (+5.09%)	:arrow_up:
torchnlp/text_encoders/moses_encoder.py	75% <66.66%> (-3.95%)	:arrow_down:
torchnlp/nn/sru.py	98.91% <83.33%> (-0.01%)	:arrow_down:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 9fc48f5...00f043d. Read the comment docs.