Some but not all CSI sidecar apps may want to cache information
retrieved from a CSI driver. Those (and only) those apps needs a way
to detect that the cached information became invalid due to a CSI
driver restart and then react to that. The Connect function supports
notifying the app via an optional callback which then can:
- kill the app via os.Exit
- invalidate the cached information and continue by reconnecting
- prevent reconnecting and exit the application when gRPC method
  calls fail with status.Unavailable

The default behavior is as before, i.e. the connection is getting
restablished.

This verifies that both unix:/// and absolute paths work and that the
timing behavior is as expected (wait for server, react promptly once
it appears).

connection: add tests

Kubernetes 1.13 already transitioned and the CSI sidecar containers
are in the process of following that (for example, in
kubernetes-csi/external-provisioner#219).

connection: replace glog with klog