When having several machines trying to start at the same time, polling
every one second consumes the API requests limit fairly easily.

Add a new `--hetzner-wait-on-polling` flag, similar to `--wait-on-error`
to change the time between requests.

Default value is left at 1 second to keep the current behaviour.

Signed-off-by: Iñaki Malerba <inaki@malerba.space>