This can be used to combine e.g. the platform's SecureRandom
implementation with our own, so that a weakness in either source doesn't
harm security as long as the other source is strong.

This should make it easier for people to build the project with a freshly
downloaded ADT Bundle.

Two changes have been made to Tor:

1. Set can_complete_circuit to false when the network is disabled, and
don't try to build introduction circuits while can_complete_circuit is
false. This avoids a situation where Tor tries to build introduction
circuits as soon as the network is re-enabled, all the circuits fail,
and then Tor waits 5 minutes before trying to build more.

2. Added a FORGETHS command to the control protocol which clears any
cached client state relating to a specified hidden service. This can be
used to flush state that's likely to be stale before trying to connect
to a hidden service with an unstable network connection.

Support for the FORGETHS command was also added to jtorctl.

This merge reduces the number of thread pools.

We don't need two separate executors for long-running IO threads.

Tor has a controller command, TAKEOWNERSHIP, and a configuration option,
__OwningControllerProcess, that work together to ensure Tor shuts down
when the controlling process dies and/or disconnects from the control
port. By using them we can avoid creating runaway Tor processes that
have to be killed with hacks.

Fixed a bug in SimplexMessagingIntegrationTest that should've caught
this.

This ensures that if two DatabaseExecutor tasks update the database and
broadcast events, the events are broadcast in the same order as the
updates occurred.

Patched jtorctl to pass more information about circuit events to the
controller to help us debug connection problems.