If a file takes <1 second we're not getting paralllization benefits

[pcn]

In the relay daemon it may make sense to write to multiple temp files and rename them all at once in order to get parallel deliveries.

[pcn]

Probably use open_next_queue_file to open a configurable number of files, and return a file-like object (though maybe one that's just duck-type file-like enough - e.g. with a write and flush method). That method can write lines round-robin between multiple files and I can measure the speed increase there.

[pcn]

This is partially working in new_sending_mechanism, but it is broken there. In production_spooling the code prior to this work is bundled up and working.