As the user points out in
https://github.com/boostorg/redis/issues/205
wait_for_all might be the cause of delay in the
reconnection. I could not observe any improvement in my
local setup but wait_for_one_error look the correct token
and all tests pass so I am willing to change it.
Before these changes the request had to be marked as written in order to
interpret incoming responses as belonging to that request. On fast
networks however, like on localhost and underload the responses might
arrive before the write operation completed.
