mesa/main: use p_atomic_inc_return instead of locking

There's no good reason for using a mutex here, as we have a simpler primitive; atomic integers. So let's use that instead, to simplify things a bit.