Fixups: post-merge updates related to PR #6241

[jayaddison]

This pull request includes a

• Bug fix
• New feature
• Translation

The following changes were made

• Instead of referring to (QUnit) test in an integration test added by Fixup: unselection of items from AJAX data sources with non-string identifiers #6241, use the fully-qualified reference QUnit.test instead, as becomes standard after Dependencies/CI: updates to prepare for jQuery 4.0.0-beta testing #6334.
• Add some guard conditions to selection:update event handlers in the integration tests, because these are not isolated from each other at test runtime, and otherwise may delay/timeout unrelated test cases.
• Regenerate distributable content.
• Regenerate NPM lockfile.

If this is related to an existing ticket, include a link to it as well.

Relates-to / merge-resolution-for #6241 and #6334.

Edit: add additional fixup descriptions.

[jayaddison]

@kevin-brown something about the searching tags does not loose focus integration test seems incompatible with #6241 -- I've commented it out here, and that allows the test suite to pass, but it's not an approach I like so I've moved this into draft status until the problem is figured out.

[jayaddison]

The problem seems to be that selection:update event handling in the integration tests isn't isolated per-test-case. Some ideas to resolve this:

• Add some kind of per-test teardown that removes relevant event handling/selection-update resources (listeners, DOM contents, internal state?).
• Attach once-only event handlers in each of the relevant integration test cases.
• Add logic to ensure that each event-handler becomes essentially a no-op despite remaining attached during unrelated integration tests.

I attempted to add a QUnit.reset handler from here to achieve the first; but this did not seem to work.

The second approach doesn't appear straightforward either; there is no .once(...) function available for event-listener attachment on Select2 instances, nor is there a way to pass {once: true} as a parameter to .on(...).

...so I've slightly-reluctantly implemented the third approach. It doesn't genuinely provide any meaningful runtime isolation between the integration test cases, but it ensure that each event listener should only perform any actions within the scope of its parent test case.

[jayaddison]

@kevin-brown it's not ideal code (details in my previous message if you're curious), but I believe this is ready for merge.

[kevin-brown]

I'll take the "temporary" hack for now. Thanks for looking into this and documenting your findings!