Corrected GCM counter incrementation to use only 32-bits instead of 128-bits

Using 32-bits has the possibility to overwrite the IV in the first 12
bytes of the Y variable.

Found by Yawning Angel
diff --git a/library/gcm.c b/library/gcm.c
index c91598c..68b6611 100644
--- a/library/gcm.c
+++ b/library/gcm.c
@@ -263,7 +263,7 @@
     {
         use_len = ( length < 16 ) ? length : 16;
 
-        for( i = 16; i > 0; i-- )
+        for( i = 16; i > 12; i-- )
             if( ++y[i - 1] != 0 )
                 break;