[PATCH weston v3] input: don't send to clients key events eaten by bindings

[Bill Spitzak]

Hate to be a pain about this, but the description of the bug (with 
Mod+x) can be reproduced and has nothing to do with whether the 
compositor shortcut was triggered. Try Mod+y (or any other key not 
consumed by the compositor) in an xterm.

You can argue about whether the 'x' or 'y' should repeat after the Mod 
key was released, but there is absolutely no reason the result should 
differ depending on whether the shortcut went to the server or the client.

I therefore still believe that this proposed fix is a serious blunder 
and should not be done. The correct fix is for clients to get an 
accurate key-down map whenever they have keyboard focus. This proposed 
fix instead breaks it in the case that it works (when the client gets 
focus) instead of the correct fix of doing something when the shortcut 
does not cause a focus change.

If clients respond in unexpected ways when keys change state than this 
is a bug in the client and needs to be fixed there.