[FIX] settings: Do not clear schema_only settings on close_context

[ales-erjavec]

Issue

Since gh-2301 schema_only settings are destructively cleared when used in combination with a context handler. This breaks (what I would hope to be) a valid use case

class A(OWWidget):
    ...
    bar = Setting(None, schema_only=True)
    def __init__(self):
        if self.bar is None:
             self.bar = ...  # not loaded from a saved workflow

    def foo(self):
        self.closeContext()
        self.openContext(...)
        assert self.bar is not None

This state where bar is None would then be saved in a workflow.
This pattern is used in Textable add-on for issuing uuids to widgets

The same pattern is also used in OWBaseLearner widget for the learner_name but without a context handler.

I am not sure what the intent there was. Presumably the semantics of the selection restore in scatter plot, ... is such that it is only applied once when restored from a saved workflow.

Description of changes

• undo changes to settings from [ENH] OWMDS, OWLinearProjection: save selection in workflow #2301
• change selection restore in scatter plot and linear projection accordingly

Includes

• Code changes
• Tests
• Documentation

[codecov-io]

Codecov Report

Merging #2691 into master will increase coverage by <.01%.
The diff coverage is 86.11%.

@@            Coverage Diff             @@
##           master    #2691      +/-   ##
==========================================
+ Coverage   75.83%   75.83%   +<.01%     
==========================================
  Files         338      338              
  Lines       59545    59566      +21     
==========================================
+ Hits        45156    45173      +17     
- Misses      14389    14393       +4

[kernc]

Presumably the semantics of the selection restore in scatter plot, ... is such that it is only applied once when restored from a saved workflow.

What if not only scatter plot; would all widgets need a self.__did_restore_state?

[ales-erjavec]

What if not only scatter plot; would all widgets need a self.__did_restore_state?

Presumably, yes.

[ales-erjavec]

A kingdom for a saveState slot.

[astaric]

@ales-erjavec the addition of sparse support in scatter plot created conflicts with this. Can you rebase?

[ales-erjavec]

Ok.

[astaric]

Thanks