Support GKE E2E tests and use (actual) E2E tests to gate releases

[rainest]

What this PR does / why we need it:

• Adds GKE and existing cluster support to E2E tests.
• Uses the E2E tests as the release job condition for creating the release tag not yet, currently just a test variant because I can't easily test with the release workflow.

We need this because we aren't currently running the test suite we want to during releases. It also requires some refactoring of the E2E tests themselves, which is a good opportunity to refactor them and KTF to address some ease of use issues. Tests cannot easily load cluster-specific functionality (loading images, adding a KIND configuration, etc.) without abusing the environment addon functionality (we have addons that fail if loaded on an unsupported cluster, which we don't want). Kong/kubernetes-testing-framework#403 seems an appropriate means of allowing for this.

Which issue this PR fixes:
Part of #2790
Fix #2287 (I think this actually got fixed earlier, but whatever, it's still open and this obsoletes it).
Fix #3041

Special notes for your reviewer:

• https://github.com/Kong/kubernetes-ingress-controller/actions/runs/3201260283/jobs/5229046505 uses the temporary E2E addition. We may want to make this permanent, but it's not the release job this eventually intends to be part of. I left it in for Add actions to run e2e/integration on k8s clusters of cloud providers #3041
• This draft only adopts the new system for TestDeployAllInOneDBLESS. Pausing for any "wait, don't do it this way" comments before doing it for the rest.
• TestWebhookUpdate requires an odd hack that's not useful anywhere else, and the changes to handle image loading generically don't help its hack. As it's an E2E test only because we can't write an integration test for it, I don't think it's worth supporting other platforms for it at the moment, and just made it KIND-only.
• The image loader still has bits of a bad design from when I originally wrote it. Remove error from image loader builder kubernetes-testing-framework#440 plans to remove this, but there's a functionally identical workaround in place here to avoid releasing a new KTF yet.

[rainest]

Discovered fun facts about environment notifications and the current E2E test job layout. 500+ low-content notifications in the timeline, very useful.

https://github.com/Kong/kubernetes-ingress-controller/actions/runs/3424207046/jobs/5703617497 is the latest run with the changes to accept and use a (single value) matrix. This also mitigates an unrelated bug in the Makefile that I can't and don't really want to figure out:

16:16:38-0800 esenin $ grep -A1 ^test.e2e.gke Makefile
test.e2e.gke:          
	CLUSTER_NAME="e2e-$(uuidgen)" \

16:16:43-0800 esenin $ CLUSTER_NAME="e2e-$(uuidgen)"; echo $CLUSTER_NAME
e2e-69ba2dd0-0578-415c-8e7c-638351a73d22              

16:16:47-0800 esenin $ make test.e2e.gke                           
CLUSTER_NAME="e2e-" \  
	KUBERNETES_CLUSTER_NAME="" go run hack/e2e/cluster/deploy/main.go \
	KONG_TEST_CLUSTER="gke:" \

I originally copied the target from the existing GKE target, which appears to be broken. uuidgen output apparently disappears into the void when make runs it, and that variable isn't being copied properly to boot. I've added a name generator to the hack code because I don't much want to think about bash, and the names are mostly just UUIDs anyway.