sanity check for rcutils_steady_time_now

[serge-nikulin]

rcutils_steady_time_now currently trusts the OS specific implementations to provide truly monotonic time values. Issue #86 and Internet show the monotonicity claim should be verified.

[mikaelarguedas]

Thanks @serge-nikulin, we were indeed using "true" monotonic clocks in the past and changes the behavior to match what we get from chrono at the c++ level. linking to the relevant issue here for book-keeping: #43

[serge-nikulin]

@mikaelarguedas, because of non-monotonic Windows steady timer (due to overflow in previous ROS2 implementation, fixed by #87) I had a silent failure in a publisher's executor.

This fix forces the system to non-silent error state in similar failure mode on all platforms.

[clalancette]

I'm going to close this as sort of a duplicate of #43. If you disagree, feel free to reopen (but I think we need an explicit actionable thing to do here).