TopicModeling: Change output for topic vizualization

[ajdapretnar]

Issue

Partially implements #450.
Partially implements #337.

Description of changes

Output All Topics is changed to output topics as rows and words as columns. Additionally, marginal topic probability is added, so topics can be visualized in MDS, similar to pyLDAvis.

While fixing this, it was necessary to fix the corpus output, which output as many topics as requested by the user, even when gensim wasn't able to find that many topics. Now the output matches the one from gensim.

Includes

• Code changes
• Tests
• Documentation

[codecov-io]

Codecov Report

Merging #483 into master will increase coverage by 0.06%.
The diff coverage is 95.83%.

@@            Coverage Diff             @@
##           master     #483      +/-   ##
==========================================
+ Coverage   61.96%   62.03%   +0.06%     
==========================================
  Files          59       59              
  Lines        6197     6216      +19     
  Branches      808      811       +3     
==========================================
+ Hits         3840     3856      +16     
- Misses       2221     2222       +1     
- Partials      136      138       +2

[PrimozGodec]

When I connect on the input corpus with election-tweets there is the dimension mismatch:

Traceback (most recent call last):
File "/Users/primoz/PycharmProjects/orange3-text/orangecontrib/text/widgets/owtopicmodeling.py", line 234, in on_result
self.Outputs.all_topics.send(self.model.get_all_topics_table())
File "/Users/primoz/PycharmProjects/orange3-text/orangecontrib/text/topics/topics.py", line 141, in get_all_topics_table
metas=np.hstack((names, topic_proba)))
File "/Users/primoz/miniconda3/envs/orange/lib/python3.7/site-packages/numpy/core/shape_base.py", line 340, in hstack
return _nx.concatenate(arrs, 1)
ValueError: all the input array dimensions except for the concatenation axis must match exactly

[PrimozGodec]

It looks good to me.