Block Editor: Maintain selection when multi-selection toggled

[aduth]

Extracted from #15927
Previously: #14640 (comment)

This pull request seeks to revise the behavior of the selection reducer to persist selected block through (multi-)selection toggling. Without these changes, an image will become unselected when resizing it, or as in #15927, the Columns block would become unselected when resizing columns.

Testing Instructions:

Verify that resizing an image does not unselect the block.

1. Navigate to Posts > Add New
2. Insert an image block
3. Select an image from the media library
4. Resize the image
5. Note that the image block is still selected

[jorgefilipecosta]

Worked well on my tests and the change makes sense 👍

[aduth]

@jorgefilipecosta In some final review of the changes, I was looking at the initial state object and it occurred to me that if the intent with "toggling" selection is to allow/disallow multi-selection, we should cancel any multi-selection (isMultiSelecting) when disallowing. I've added this in 0be9fea.

[aduth]

A few notes on my documentation attempts here (cc @ellatrix):

• I assumed BLOCK_SELECTION_EMPTY_OBJECT was used here to allow for some strict equality checks in selectors, though I could not find any place we actually do this. The tests also pass if its use below in BLOCK_SELECTION_INITIAL_STATE are replaced with new objects. Can you clarify what purpose this value serves?
• And on that note, for what reason does it need to be an object? I had planned to document start and end as an object, but could only find that we reference clientId on it. Are there other properties, or could it just be replaced with a direct reference to the clientId?

[ellatrix]

Don't we do any strict equality checks in the reducer? Perhaps I removed that later on, so this wouldn't be needed any longer.

The selection object can also include the rich text identifier and caret position, if a rich text instance is selected.

[paaljoachim]

@jorgefilipecosta @aduth What is still needed to get this merged?

[youknowriad]

Up again here :) should we close or land this PR?

[aduth]

It appears this is no longer an issue. I cannot reproduce the original steps.

I suspect it was fixed separately as a result of #17467.