methods to help manipulate and update DocumentTermMatrix incrementally

[tanmaykm]

This adds a few methods to help manipulate and update DocumentTermMatrix incrementally:

• serialize and deserialize: optimized by not serializing the column index unnecessarily - that is re-constructed from the terms vector upon deserialization
• prune!: removes documents and terms from an existing index - those that would correspond to deleted documents
• merge!: merges two instances - one being the incremental update to be applied on an existing full index

Also relaxed StatsBase version requirement to include both 0.32 and 0.33. This was preventing this package being used together with certain other packages.

fixes #243

[tanmaykm]

I see now that copyto! for sparse matrices is quite slow for large matrices. It should be possible to implement a different copy method optimized for the specific needs of merging DTMs in TextAnalysis. I shall send another PR with that in a bit.