Suggestion - not an issue as such

[lulu-2021]

I am working with very large volumes of data and since the IDF has to relate to the whole document space - it is difficult with your version to load the entire document space.. Based on your concepts I have re-worked the code to use streams and lazy eval for a large corpus and also OTP based parallelisation.. Once i am happy with the code test wise I am happy to submit a PR - if you think this would add value to your version - otherwise I will build my own :)