add exec_on test helper

[rade]

so we don't repeat the remote docker invocation command (complete with hard-coded port) over and over again.

[rade]

Note that there is a bug here... there are a bunch of tests that used to run as sh -c. I stripped that off. But now the error handling doesn't work, i.e. the scripts just terminate at that point. Alas it turns out that was the case before too! I experimented with a few variations but no luck. @awh I welcome suggestions on how to fix that. I wonder whether we should just test the exit status, and dispense with the "&& echo ok".

[awh]

I encountered this yesterday - I think it's caused by the set -e at the top of config.sh, which is sourced into all the test scripts. As a consequence a statement such as

ok=$(docker -H tcp://$HOST1:2375 exec c1 sh -c "ping -q -c 4 $C2 >&2 && echo ok")

would cause the script to exit abruptly if the ping fails due to it returning a non-zero status. I'll propose a solution as part of the work I'm doing to make run_all.sh behave nicely in the case of test failures (see #591).