Bug fix: XOutputDevice.RefreshInput() is not thread safe.

[kenh6942]

Originally observed issue:

• When moving 2 joysticks on 2 different input devices mapped to 1 XOutput device at the same time, occasionally one of the joystick inputs would be lost/disconnect.
• In game, this caused the joystick to remain pressed in some last-polled direction.
• This could be resolved in the XOutput app by hitting the "Force Refresh" button.

Debug/conclusion process:

• Log would show a "WARN Poll failed for ..." when this occurs.
• Log event is from DirectDevice.cs line 319 (logger.Warning($"Poll failed for {ToString()}");)
• The exception thrown to cause the log event was a "Collection was modified ..."
• Exception was thrown from line 313 (InputChanged?.Invoke(this, deviceInputChangedEventArgs);)
• ...which was invoking XOutputDevice.SourceInputChange() which then calls RefreshInput()
• Debugging calls in the function, it appeared to be the call to changes.Any() that was throwing the exception.
• Looking online for other instances of when such an exception could occur, the quick solution is to make a copy of the Collection for iteration. Adding ToList() to line 119 (changes.GetChanges(force);) appeared to fix the issue, but is more masking the problem than fixing it.
• How is the Collection being changed? Each DirectDevice object that eventually calls into XOutputDevice.RefreshInput() has a thread. RefreshInput() makes changes to the state variable of XOutputDevice. It is possible for more than one thread to be calling RefreshInput() and thus making asynchronous changes to the state variable.
• Excluding it's initialization, state is only used/of-use in RefreshInput().
• Solution: Make state a local variable of RefreshInput().

This does mean a DeviceState is instantiated for each call to RefreshInput() which is inefficient.

[kenh6942]

Originally branched off the 3.32 tag. The forced push was to rebase on top of 3.x branch.