What?

I know changing block for performance is generally frowned upon, but hear me out:

• The paragraph is special because it's unusual to add an unconditional useSelect call for a control. Normally controls don't require a useSelect call and simply rely on attributes. Most of the time it's block supports hooks that do this. (I did wonder if this should become a hook that's only supported by paragraph.)
• The paragraph is the most frequently used block. While I don't think we should be over optimising (like the pure thing), I don't think avoiding the store subscription is because it's a paragraph.
• We have lots of pecedent for:
  • Creating wrapper components for controls: even in the paragraph we have ParagraphRTLControl.
  • Moving useSelect store subscriptions within components mounted in InspectorControls or BlockControls so it's only for the select block: for example list item's IndentUI.

Why?

Every useSelect added to a (frequestly used) block will degrade the load and type bit. By moving it within InspectorControls, the subscription is now only added for the selected block(s).

How?

Moving it with InspectorControls ensures it's only added when the block is selected.

Testing Instructions

Load (first block)
First run: -4.25%
Second run: -3.85%

Testing Instructions for Keyboard

Screenshots or screencast