use kubectl replace --force for one-off pods

[balopat]

when a "one-off" (short lived) pod is defined with RestartPolicy: Never, the following happens with skaffold dev:

1.) image is built
2.) kubectl apply succeeds - pod is deployed, pod completes
3.) code change triggers loop
4.) image is built
5.) kubectl apply succeeds - pod does not restart, skaffold just says "Waiting for changes...", there are no logs, and the pod will still have the old version of the image

Proposed behavior:

1...4.. same
5.) kubectl apply succeeds, skaffold recognizes that pod's image is not the same tag as the latest build - prints warning, and runs a kubectl replace --force instead. Pod gets deleted and redeployed, skaffold streams new logs.

Open question(s):

• should this be only for skaffold dev or skaffold run as well?

cc #891
cc #940
cc @coollog

[coollog]

I think it makes sense for skaffold run as well since a successful skaffold deployment should probably give the promise of the intended running pod(s) using the image built during the build stage.

[balopat]

With the --force flag this is covered I believe.