Sparsity of data when values come from compute values

[nikicc]

Orange version

master

Expected behavior

When compute values produce sparse features (e.g. BoW), X array should become sparse not matter in which format (sparse/dense) it previously was and regardless of number of features (zero, non-zero) it previously had.

Actual behavior

Sparsity of X, Y and metas is determined from the sparsity of source data, regardless of the sparsity of data coming from compute values.

Steps to reproduce the behavior

Most easily on text. Learn a classifier on one BoW data and apply it on an other data that initially has some features.

Additional info (worksheets, data, screenshots, ...)