Remove wrappers and other cruft

[mpenkov]

Removed gensim/models/wrappers, associated tests and documentation entirely.

Removed gensim/sklearn_api entirely.

Removed simserver.rst - seems like something really ancient (e.g. https://radimrehurek.com/gensim/simserver.html).

@piskvorky Some questions:

• Is there anything else worth removing?
• What should we do about the coherence model tests? They are useless without the wrappers.

[piskvorky]

Thanks! Can you add a summary into the Migration wiki too please? So I don't forget to describe it.

[piskvorky]

Aaaah, here it is. This PR wasn't labeled as "Milestone 4.0" nor linked to its issue #2852, so I couldn't find it. But I knew there was more :)

Is there anything else worth removing?

Yes, for sure. My dilemma is that I'm not sure whether I dislike some modules simply because I never used them (~NIH), or because they're objectively useless / poor.

Although my intuition leans heavily toward the latter.

Problematic modules by their mailing list activity:

• Group 1: Desirable but weak execution. Top candidates that people report issues with (=try to use and fail) are HDP, summarization, dtm, pattern and the mallet wrapper: So either fix / implement properly (unrealistic for us unfortunately) or remove those.
• Group 2: Unclear. More obscure wrappers with occasional reports are tensorflow/keras, sklearn, wordrank. Remove?
• Group 3: Never-heard-from-users. Pivoted normalization, poincare, vowpalwabbit, simserver, ?more? (not top of my mind precisely because never discussed). Remove.

What should we do about the coherence model tests? They are useless without the wrappers.

I'm not sure, would have to take a closer look.

CC @gojomo any candidates from your POV?

[mpenkov]

From my point of view, the problem is that I don't have a full understanding of what needs to go and what needs to stay. Once that gets decided, I can execute relatively quickly.

Of course, making the big decision is the hardest part of this task.

[piskvorky]

OK. Not blocking for 4.0.0beta.