taint node with PreferNoSchedule to prevent receiving (and double draining) additional pods from other rebooting nodes

The default of 0 is taken as "delete immediately", which is
not appropriate.

Without this, golint is complaining about a few cosmetic changes.
This solves it, and is necessary if we want to add a lint test
in CI.

Moving the drainer object close to its usage is more readable.

	This is #152, #139, #127 in disguise.

	Maybe this time let it simmer a bit longer until the k8s
	release is there?

	This is a follow-up to #150, so we can get a 1.4.x release
	out that will be geared towards k8s 1.1[6-8].

	Update to latest 1.17 kubectl: 1.17.7.

	After the release of kured 1.4.0 we should be able to go back.

	This was decided in our meeting
(https://docs.google.com/document/d/1bsHTjHhqaaZ7yJnXF6W8c89UB_yn-OoSZEmDnIP34n8/edit#heading=h.8cgszb6vuhza)

	Let's go with supporting 1.1[678] in this release.

	confirmed by running https://github.com/kubernetes-sigs/clientgofix

	closes: #123

Use the tools installed in the host to effect reboots, and allow
the execution of commands such as `needs-restart` to determine if
reboots are required.

Fixes #12

`kubectl drain` now ignores undrainable things and waits for
termination, so we no longer have to do that ourselves.

Zero-valued alerts are now ignored, and the names of the first ten
blocking alerts are logged for diagnostic purposes.

Test for schedulability immediately prior to each reboot check instead
of once at startup, thereby accomodating external changes to this value
during the lifetime of the daemon.