Restore wasn't fully implemented anyway, and having it enabled would
occasionally cause macOS to report "Bluetooth unavailable" after
reconnecting BLE sessions several times.
After turning that off, macOS treats implementing
`WillRestoreStateEvent` as an error, so I removed that.
We really use the socket state to control the session, so there's no
need to also pass a CancellationToken around. The exception was being
caused when an unlucky callback happened after a session was disposed
and the callback tried to get a new token to check cancellation. If the
token source is already disposed, it can't provide a new token.
Also, remove some excessive logging and add more exception logging.
An `EventAwaiter` timeout now throws a `TimeoutException` instead of a
`TaskCanceledException`.
Disconnecting a micro:bit while subscribed for change notifications was
causing an exception because a notification was being triggered after
the socket closed. Fixing that led to an attempt to close the socket
after it was closed and set to `null`. That one was solved with a null
check.
For BT, connection is still not reliable but it's a step in the right
direction... I think. The socket lock is now held while opening the
RFCOMM channel, which is an awful long time to hold a lock, but really
no other thread should be messing with the socket during that time so I
guess it's the right thing to do. I also added debug printouts on the
RFCOMM event delegate.
