Repo/test manifest divergence

[rainest]

What this PR does / why we need it:

Adds CI task to run the manifest generation script, check if it resulted in changes, and fail if it did. This ensures that single and base manifests do not diverge.

Which issue this PR fixes (optional, in fixes #<issue number>(, fixes #<issue_number>, ...) format, will close that issue when PR gets merged): fixes #668

Special notes for your reviewer:
https://github.com/Kong/kubernetes-ingress-controller/runs/960118920?check_suite_focus=true failed because it should. The manifests aren't in sync currently. In a local test directory where I have updated them:

15:20:02-0700 yagody $ git status
On branch repo/test-manifest-divergence
nothing to commit, working tree clean
15:20:04-0700 yagody $ ./hack/build-single-manifests.sh 
15:20:21-0700 yagody $ git status                      
On branch repo/test-manifest-divergence
nothing to commit, working tree clean
15:20:22-0700 yagody $ make verify-manifests
./hack/verify-manifests.sh
diffing ./hack/../deploy/single/ against freshly generated single manifests
./hack/../deploy/single/ up to date.

How do we want to handle that? Just update them and roll the changes into this PR, or separate PR to update them and then rebase this onto next (or main) after it's merged?

[mflendrich]

Looks like this check is failing at introduction 😁

[mflendrich]

Ah, should have read Special notes for your reviewer first. 🤦

How do we want to handle that?

I recommend a TDD-like approach here:

1. introduce a failing test
2. fix the cause of failure
3. anticipate test success 🎉

For me, this can (or even should) happen in 1 PR. It does not feel right to deliberately introduce a CI failure. I won't insist, though.

[rainest]

Yeah, mostly a question of just where that change goes, which is a minor procedural point and not particularly important.

Given that it's pretty minor and I have the link to an existing test run demonstrating the failure already, going ahead and pushing the manifest updates within this PR.

[rainest]

Latest updates apply @mflendrich 's bash suggestions and comingle a manifest update along with the CI update so that the test now passes. Please refer to the run linked earlier/CI history for a demonstration of the failure case behavior.