[Windows] Fix left/right Shift key regression

[SupSuper]

This fixes a regression introduced in #92415 and reported in #101384.

When multiple Shift keys are pressed, Windows only sends a KEYUP for the last key released, leaving the other stuck pressed. A workaround was introduced in #80231 to fake this missing KEYUP event.
However this workaround was broken in #92415. _get_mods() returns the modifier keys before the KEYUP, so there will always be a Shift key pressed before it's released.

This PR restores the original check and respective functionality. The MRP in #101384 can be used to reproduce and confirm the fix.

Thanks to @ghostsoft @andyprice @romlok for helping debug the issue.

• Bugsquad edit, fixes: Keys do not release properly if left shift and right shift are held at the same time #101384

[Zireael07]

Can you add a comment explaining WHY this function and not the other?

[bruvzg]

The difference between mods/GetKeyboardState and GetAsyncKeyState is how they interact with message queue. For a fake key injection in the raw input handler, it is likely correct to use GetAsyncKeyState since normal input messages are not processed at this moment. Will test it tomorrow.

[SupSuper]

@Zireael07

Can you add a comment explaining WHY this function and not the other?

The goal is to send a fake KEYUP event when a Shift key is released after a Shift key is pressed. (i.e. check if Shift keys are still pressed after the KEYUP).
mods/GetKeyboardState doesn't work for this because it returns the state before the KEYUP is resolved, so there will always be a Shift key pressed. You either end up sending no fake events or duplicate events.

The only solutions were:

• Extend _get_mods to also handle Left/Right modifier keys so we can check for both pressed.
• Rewrite everything to use RawInput.
• Use GetAsyncKeyState to check the real key state (the old solution).

Since the old solution still worked, I opted for the simplest fix.