Bring back applying labels to services deployed with helm

[tejal29]

Fixes #2379

In #965, we turned off labelling services due to kubernetes/kubernetes#68087
This bug is in the kube-controller-manager which runs on the cluster master side.
I tested this PR on my cluster with 1.11.7-gke.4 and port-forwarding for LoadBalancer type services works.

Impacted versions for this bug

GKE cluster Version	State	Notes
v1.12.0 and onwards	Never Occurred	As per the Changelog for v1.12.0 here, this bug never appeared in 1.12.0 . The very first version lists kubernetes/kubernetes#68087 being fixed in notes. See L1243
v1.11.x	Occurred in v1.11.3 and before	This fix was patched in v1.11.4 as per the Changelog for for v1.11.0 here. As per release notes on Jan 2019, all cluster masters running 1.11.2 through 1.11.4 will be upgraded to 1.11.5-gke.5. which should contain the fix.

As per GKE version skew policy, June 2019 release notes mention GKE will only support the three newest minor versions.
So in june end, for all node pools older than v1.11.X Google will enable node auto-upgrade and these nodes will be updated to 1.11.10-gke.5. ( we are v1.14 now)

In conclusion, i feel this is safe way to enable port-forwarding for services deployed by helm (helm2) to get in.

[codecov]

Codecov Report

Merging #2568 into master will increase coverage by 0.89%.
The diff coverage is n/a.

Impacted Files	Coverage Δ
pkg/skaffold/deploy/labels.go	16.88% <ø> (+0.83%)	⬆️
pkg/skaffold/runner/build_deploy.go	65.9% <0%> (-2.35%)	⬇️
pkg/skaffold/util/util.go	85.91% <0%> (-0.76%)	⬇️
pkg/skaffold/docker/image_util.go	0% <0%> (ø)	⬆️
pkg/skaffold/jib/jib_init.go	89.39% <0%> (ø)
pkg/skaffold/sync/kubectl/kubectl.go	9.67% <0%> (ø)
pkg/skaffold/runner/context/context.go	62.5% <0%> (ø)
pkg/skaffold/debug/debug.go	45% <0%> (+1.66%)	⬆️
pkg/skaffold/docker/client.go	76.29% <0%> (+2.55%)	⬆️
pkg/skaffold/build/local/types.go	71.42% <0%> (+2.85%)	⬆️
... and 8 more

[priyawadhwa]

LGTM -- should we log a warning if the kubernetes version is lower than 1.11.3 and we are deploying with helm? Since we do that for kubectl version already.

[balopat]

I wouldn't add a warning - if someone sees the issue, we should just ask them to upgrade to 1.11.4+