execCommand should support specifying timeout

[joel-bluedata]

Ideally we can model the timeout as an input to this function; if not we need a usual way for callers to model a timeout and interrupt it.

Related to issue #54 (as a general backstop against never-terminating startscripts).

[joel-bluedata]

Pulling this into 0.5.1 as a possible solution for #213 if we can figure this out.

[joel-bluedata]

I think the only real way to do this will be to run an ExecCommand invocation, or maybe just the final exec.Stream, in its own separate process in the container. In general golang has a real problem dealing with a goroutine that is indefinitely blocked on I/O.

[joel-bluedata]

This is a nasty one but we need to keep returning to think about it.