Avoid overwriting label when re-initialising Deployment object

[mikery]

When I run the first example using the latest version on pip (0.4.2), the Nginx test fails with:

>       assert len(pods) == 3, 'nginx should deploy with three replicas'
E       AssertionError: nginx should deploy with three replicas
E       assert 0 == 3
E        +  where 0 = len([])

Some debugging showed that _add_kubetest_labels() is called twice: once by the applymanifests decorator, and once during deployments.get_deployments(). This results in the Deployment and Pods having differed UIDs, which causes get_pods() to return an empty list.

This patch fixes the problem by first checking to see if the deployment already has a UID label in its metadata. If so it is retrieved from the Deployment's label, instead of overwritten by a new UID.

There are already some PRs relating to this issue (e.g. #142), but label-based matching seems like a nice solution to the problem and disabling it entirely might lead to unpredictable results in more advanced multi-deployment test cases.

I lack sufficient familiarity with the codebase to know if this is the correct fix, but it is enough to get the example tests working as expected. I only tested deployments but statefulsets/daemonsets share the same code so presumably have the same issue.

[marcoceppi]

Hello, thank you for submitting a patch to kubetest! I've performed an initial scan of the changes and don't see any nits in formatting that might hold up a review. As such I've assigned @edaniszewski to take a look at this for a technical review

[edaniszewski]

This makes a lot of sense! Thanks for debugging and contributing your solution (: