Prevent axis-based actions from getting stuck

[KoBeWi]

Fixes #81164

Not sure how much correct is that. This assumes that only InputEventJoypadMotion has this problem and that you have only single axis event assigned to an action (which is true in 99.9% cases). Also a special handler for one event type feels hacky.

[Sauermann]

This is difficult to review, because I cant test the PR.
I'm not sure if there is a better way to fix the bug, but I can follow the argumentation of the shortcomings in the OP and understand the changes.

So this is a "it will not worsen the situation" review approval.

[Calinou]

This doesn't appear to work correctly on my end (using keyboard and wired DualSense on Linux at the same time). In fact, I can still notice #45628 happening on my end (look at times when I press keyboard keys while keeping the gamepad axis still):

simplescreenrecorder-2023-08-30_23.27.10.mp4

CheckBox indicates state of is_action_pressed(), left ProgressBar is the action strength, right Progressbar is the action raw strength.

Testing project: test_analog_and_digital_input_2.zip

[KoBeWi]

Fixed. It was the same issue - the action was released multiple times in a row.
Thinking again, the issue makes sense. Axis actions are pressed based on the deadzone, so if you receive multiple events above or below deadzone, the action state will change abnormally.

[Sauermann]

Origin for this PR is #80859, which originally intended to fix #45628. Since I can't replicate that issue, I can't verify that.

#80859 introduced #81164 which this PR intends to fix. I can confirm #81164 in v4.2.dev4.official [549fcce]
With this PR applied to v4.2.dev.custom_build [fb4edf5], I tried for over a minute to cause, #81164 but was unsuccessful to do that. So I am confident, that this PR fixes that issue.

Comment #81170 (comment) mentioned a problem, that a previous version of this PR introduced and I have tested and can confirm, that this has been fixed in the latest update.

The concerns of my comments regarding the implementation also have been solved.

[akien-mga]

Seems to work fine, and I don't see any regression on the joypads demo.

[akien-mga]

Thanks!