Improve netstack scalability

Netstack today uses a protocol goroutine per TCP endpoint. This creates an issue as the number of active endpoints increases. The delivery of inbound packets causes a large number of interactions with the go scheduler as the goroutines are scheduled/descheduled as they process incoming packets.

This bug is to track work to see if we can move away from this model to a mode where for most inbound traffic there is very little scheduler interaction.