Fix Input::remove_joy_mapping

[MJacred]

Erasing a joypad mapping can invalidate other attached joypads and the fallback mapping guid

Fixes #91257 (cannot confirm crash on Linux. Untested on Windows where it was observed)

If this PR is merged, it needs to be cherry-picked/backported to all supported Godot versions, afaik.

NOTE

• I'm not sure if I handled fallback_mapping correctly, because according to Input::is_joy_known the fallback does not count as a known device… Therefore, feedback on this would be nice
  • If it's only meant as an internal fallback, a more minimal change of this PR to the fallback logic would be to deny removing the mapping for that one. Then I could remove the setter and getter for the fallback. -> Removed bindings
• I renamed some members in windows joypad file to differentiate between direct INPUT and direct X. I found d_joypads, and it feels like the d_ prefix was meant as a direct input abbreviation. So I used that. I'm open to alternative naming convention, though. Some renaming should happen to untangle INPUT and X. -> Renaming is done in another PR

[MJacred]

status update:

• Rebased PR to latest master, resolving merge conflict.
• Removed bindings in Godot's spirit of "don't give bindings to things people don't need"

[akien-mga]

CC @Rindbee if you can review this as you're also familiar with that code after #95486.

[Rindbee]

I find it easier and more intuitive to record the old index than to calculate it. There is nothing wrong with the logic.

[Rindbee]

@MJacred Things are more complicated than I initially thought. Especially in the case of being between the minimum removal index and the maximum removal index. For example, suppose that more than three indexes are removed, and there is another guid between the minimum removed index and the maximum removed index, and the guid has multiple mipping records.

--------r0--ai-----r1-aj---r2----r3------

Using simple variables is not enough to record these shifts. To do this, we may need to record all of the mapping indexes of the p_guid.

[MJacred]

@Rindbee: I think I got it.

Alternatively, I can iterate over joy_names in the first for-loop, right where I remove the entry and fix the offset for each affected joypad right then and there

EDIT: fixed syntax check failure and typo

EDIT 2: did a comparison with original commit, and the differences are

• early exit if nothing removed
• use fallback_mapping instead of setting to -1 for joypads that used this guid and which was not the fallback mapping
• simplified joypad mapping fix if (count == 1) (which should be the case 99% of the time)
  • else, the old inner-loop is back again
• 2 new helper variables allowing the early exit and skipping unnecessary mapping fixings

[Rindbee]

There is no problem with the logic and it is relatively clear.

[Repiteo]

Thanks!