Adding support for unigram sentencepiece model

[tejasvaidhyadev]

As discussed in #44
I am adding support for the sentencepiece unigram model with DataDeps to download pre-trained vocab file with their log-likelihood
for more detail refer to the blog post here
Roadmap

• DataDeps for Downloading Vocab or .protofile
• unigram encoder algorithm implementation in Julia
• Test and doc

[tejasvaidhyadev]

• Added Datadeps dependency for downloading vocabulary file
• For now, I am hosting the vocabulary file on git.

[tejasvaidhyadev]

Hi @Ayushk4,
should I add docs in Readme?

[Ayushk4]

Travis CI fails on Julia 1.0, due to new DataDeps as dependency - requires updating Project.toml.

[Ayushk4]

@oxinabox Should we drop support for 0.7 tests now in Appveyor?

[Ayushk4]

Since DataDeps have been newly added, Travis and Appveyor also need to be updated to allow DataDeps downloads
Check how to set variables for CI here -
https://github.com/JuliaText/CorpusLoaders.jl/blob/master/appveyor.yml#L2
https://github.com/JuliaText/CorpusLoaders.jl/blob/master/.travis.yml#L7

[oxinabox]

@oxinabox Should we drop support for 0.7 tests now in Appveyor?

Yes

[Ayushk4]

Please look into setting DATADEPS variable for Travis and Appveyor CI. Also remove 0.7 Julia tests for appveyor.

[tejasvaidhyadev]

Thank you.
The suggested changes have been made and I hope PR is ready for further review

[oxinabox]

Cool just a few last things this this should be good to go.

@Ayushk4 can merge once happy with it.

[Ayushk4]

Looks good to me.