mgr/cephadm: fix host drain with agent

mgr/cephadm: fix host drain with agent

Agent was not getting removed from hosts with _no_schedule label
since it was using all hosts rather than _schedulable hosts. Added
a _non_draining hosts function to return a prope list of hosts agent
is okay to schedule on.

Was stuck between using a daemon to report which daemons are on the
host and wanting to remove all the daemons on the host when draining.
Now using ssh to confirm all the daemons have left the host.

Signed-off-by: Adam King <adking@redhat.com>