A performance-oriented test decoding 1 million messages from a stream

[tbkka]

This is inspired by a slow case discovered by the oss-fuzz service, this attempts to decode a stream consisting of a million zero bytes. Each of these zero bytes represents a single message with no non-default fields.

I'm checking this in to help aid in future performance optimization for AsyncMessageSequence.

[thomasvl]

I guess the only down side is these will slow down all test runs. But having it to eventual do the performance improvements makes some sense.

[tbkka]

I cut it back to 100,000 bytes instead of a million. That should reduce the overall impact on test time. I've factored out the number of iterations so future tinkerers can easily experiment.