improve Classification layer accuracy and area report algorithm

[Andreas Janz]

Discussed with Sebastian van der Linden.

• clearly label rows/columns as predicted and observed

‌

• in simple random sampling acc. ass., use integer instead of X.0 values
• would be nice if the user could define the mapping between predicted and observed classes:
  could be implemented via a custom parameter widget → Error Message is enought

‌

[Andreas Janz]

Done:

‌

[Andreas Janz]

resolves #934

→ <<cset 0e9fee12a1f9>>