Optimize mcasting GO message

[jbaumanTT]

Currently we use noc_async_write_multicast_one_packet to send out GO messages, which requires setting all the registers needed after waiting for workers. We should set all the registers before, so that we can trigger the mcast with a single register write to reduce latency. We can also avoid modifying most registers if we need to send the same go message to multiple multicast addresses.