Is your feature request related to a problem? Please describe.
Gensim apparently causes quite some issues on installation and in some widgets.
It is used is:

1. Corpus and Preprocessing: gensim.corpora.Dictionary
2. BOW: gensim.models.TfidfModel
3. in util: Sparse2CorpusSliceable
4. Topic Modeling: GensimWrapper
5. Topic Modeling: CoherenceModel

Describe the solution you'd like

1. Write own Dictionary (or copy from Gensim)
2. Use sklearn.feature_extraction.text.TfidfTransformer
3. Figure out Sparse2Corpus (minor)
4. Use LDA and other decomposition methods from sklearn.
5. Remove Coherence model. Apparently, there is no suitable alternative to run it with sklearn. I don't think it warrants keeping gensim just for this.

Describe alternatives you've considered
SpaCy could also be used for TM and similar.