The TCP stream of the Sink might not get closed when you expect

[dmorn]

The native RTMP client is not going to flush and close the connection at end of stream, but rather at component termination. For example, if the sink is part of a complex pipeline with multiple endpoints, it will not terminate until all components do. It might happen then (and it did to us) that the RTMP server receives the very last message from the sink but it's kept in a stale state till the whole pipeline exits, even though the finalize function was called right after the last message, at end of stream.

See #56 for more context.

Whatever the solution, it needs to be documented.