[ENH] Corpus & Bow: Improve sparsity handling according to Orange>=3.8.0

[nikicc]

Issue

BoW features when comming from compute values were dense, when some features already existed in the corpus. The core of the problems is bad logic for deciding on the sparsity, which was fixed in biolab/orange3#2341 and released in Orange=3.8.0.

Description of changes

• Remove the preference for sparse matrix when X is empty.
• Mark all BoW features as sparse — this assures that X becomes sparse even when data comes from compute values.
• This also fixes the currently failing tests on the master which failed due to updates in Orange=3.8.0.

IMPORTANT: Orange version in requirements.txt wasn't bumped to 3.8.0 on purpuse and the code was written in a backwards comparible manner so that the addon can also work with older versions of Orange. Note, however, that it is higly recommended to update Orange to >=3.8.0 since as of this PR the BoW data on older Orange versions will become dense!

Includes

• Code changes
• Tests
• Documentation

[codecov-io]

Codecov Report

Merging #281 into master will decrease coverage by 0.33%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #281      +/-   ##
==========================================
- Coverage   85.26%   84.92%   -0.34%     
==========================================
  Files          33       33              
  Lines        1866     1864       -2     
  Branches      337      336       -1     
==========================================
- Hits         1591     1583       -8     
- Misses        239      242       +3     
- Partials       36       39       +3