Cancellation on delete could cause us to stop tracking object

[hausdorff]

Following up from #244, in the process of review, I realized that in a handful of cases, it is possible that ctrl-c during a delete could cause the Kubernetes provider to stop tracking the object, even though it exists.

The reason is that provider does not actually check whether the returned value conforms to await.InitializationError, which is what the await package uses to signal that the API server has acknowledged the operation, but has not yet completed it.

The fix seems pretty straightforward:

• "Unpack" the error in kubeProvder#Delete (similar to how we do it here:

  pulumi-kubernetes/pkg/provider/provider.go

  Line 439 in 2efb314

  if awaitErr != nil {

  )
• Rename InitializationError to something more appropriate -- perhaps PartialError?

[lukehoban]

@hausdorff How serious is this in practice? Have users hit this? Do we need to make sure to fix in 0.19?

[hausdorff]

This is very minor, you'd have to get very unlucky. But, we should solve it, let's put it into 0.20