[FIX] table_from_frame: replace nan with String.Unknown for string variable

[PrimozGodec]

Issue

1. When transforming pandas data frame to table string columns that contain nan values will
   keep them after the transformation but Orange uses an empty string for the unknown value in the StringVarable.

   For example:
   df = pd.DataFrame(
   [["a", "b"], ["c", "d"], ["e", "f"], [np.nan, np.nan]],
   )

   will be transformed to the table with two string variables and nan values will be kept even String.Unknow = ""

2. When a column is recognized to be a string and has an object type, it still can contain some values that are not strings. Cast column to string.

Description of changes

Changed that nan values are transformed to String.Unknown for columns that will be transformed to the string variable and values are transformed to strings.

Includes

• Code changes
• Tests
• Documentation

[markotoplak]

Could you better explain what is the reasoning here?

Does not Orange use the empty string OR nan? I'd lean towards using nan wherever possible. And this goes further from what I'd like to see. :)

[PrimozGodec]

Could you better explain what is the reasoning here?

I added a better explanation of the issue.

Does not Orange use the empty string OR nan? I'd lean towards using nan wherever possible. And this goes further from what I'd like to see. :)

I am not sure if nan is supported by all widgets (at least it does not work in Orange3-text addon). I can start working on supporting it there. I changed my implementation that it uses StringVariable.Unknow instead of hardcoded "" such that it will still work when we decide to use nan instead of "".

[codecov]

Codecov Report

Merging #5795 (3e6f5bf) into master (c860359) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #5795   +/-   ##
=======================================
  Coverage   86.12%   86.12%           
=======================================
  Files         316      316           
  Lines       66386    66386           
=======================================
  Hits        57173    57173           
  Misses       9213     9213