Detaching CPU intensive operations from main actor threads

[stedfn]

Actix actors are single threaded, which means they get blocked by CPU intensive operations and delay the handling of other messages. Moving these operations on separate OS level threads will allow the actors to remain responsive and continue processing incoming messages concurrently while heavy computations run in parallel.

One way to detect such issues is using flame graphs. An example can be viewed here, in the PeerManagerActor thread in a forknet experiment with a padded state witness size:

Here we can notice that the handle_msg_network_request calls send_message_to_account which spends most of the time in the sign_message function, which is a CPU intensive operation since it serializes the PartialEncodedStateWitness so it should be detached from the main thread.

An example PR for the PartialWitnessActor can be found here: #12656