Reinstate detection/skipping of invalid/down/unreachable hosts

[bitprophet]

v1 had skip_bad_hosts which worked in the execution machinery to skip executions of tasks on hosts that could not be connected to. Need to revisit exactly how that worked and how we might want it to work in a 2.x context.

Seems like it would dovetail very neatly with #1808 which is about retrying connections; arguably skipping should just be a branch at the end of the retry procedure once all retries are exhausted.