Fix missing optimization in CodeSinkingPass (KhronosGroup#6703)

Set `checked_for_uniform_sync_` after the first scan for uniform memory
sync instructions in `HasUniformMemorySync`. Since this function is
called for most instructions, this change makes the optimization pass
O(N) instead of O(N^2). This drastically reduces how long it takes in
complex binaries and fulfills what I assume was the original intent of
`checked_for_uniform_sync_`.

Author: diddily

source/opt/code_sink.cpp

@@ -246,6 +246,7 @@ bool CodeSinkingPass::HasUniformMemorySync() {
     }
   });
   has_uniform_sync_ = has_sync;
+  checked_for_uniform_sync_ = true;
   return has_sync;
 }