Fix keep_n behavior for Dictionary.filter_extremes

[johann-petrak]

If keep_tokens are specified, this fixes the problem by using a lambda for
sorting which returns a fake document frequence equal to the number of documents
for all tokens in the keep list.

Fix #2230

[menshikh-iv]

Thanks for PR @johann-petrak, can you

1. Add test for this case (to avoid regression in future)
2. Merge latest develop to your branch (outdated base branch is an reason of failed CircleCI)

For last you need

• Configure remote for your fork (current gensim repo) https://help.github.com/articles/configuring-a-remote-for-a-fork/
• Merge upstream/develop to your branch: https://help.github.com/articles/syncing-a-fork/

[johann-petrak]

Thanks! Sorry, still not too familiar with pull requests in general and gensim in particular.
So - the development branch is the one which should be used to get the latest development head and for testing and fixing issues?

[johann-petrak]

I think this should be ready to get merged, please let me know if anything should be still missing.

[johann-petrak]

Sorry to bump this, but I did think I did everything necessary, is there still anything missing from my side?

[piskvorky]

@johann-petrak thanks a lot! We'll get to open source in a few weeks again (swamped now). Is merging this PR urgent for you?

[johann-petrak]

Not really urgent just wondered if there might still something missing.

[menshikh-iv]

Sorry for waiting, thanks for PR @johann-petrak 👍