AMQP transfer abort is occasionally ignored.

[kgiusti]

See Issue #242 and Issue #1288

When a transfer abort is received from an upstream AMQP sender the router must proxy the abort downstream to all outgoing deliveries corresponding to that transfer.

The implementation uses an atomic flag in the message content shared data structure.

The problem is that flag is only checked by the outgoing transfers IF more data is sent out the outgoing transfer. This may not be the case, especially since the upstream receiver no longer receives data after the abort occurs.

The abort event should be handled more deterministically. Ideally it would be represented by a delivery update event passed to the outgoing delivery.