Batched dry-run? No thank you!

[timothysmith0609]

What are you trying to accomplish with this PR?
During validation we attempt to send dry-run=server apply all resources, but there's some flaws in this approach:

• Even if the batched apply fails, we just fall back to per-resource validations, anyway, so anything caught by server-side apply does not actually raise an error, anyway
• We were not separating out applyable vs non-applyable resources (e.g. resources with generateName field). This would cause the dry-run to fail because the incoming resource can not be applied. This is easily fixed by simply only taking applyable resources, but the fact this bug has existed for so long and no one has fixed it speaks to the lack of value batch dry running is adding
• More broadly, batched dry-run is at odds with how Krane breaks up a deploy into multiple stages. It does this because some resources are, in general, dependencies of other resources in the same apply-set. E.g. deploying a ConfigMap/Secret/ServiceAccount that is mounted by a Pod. Server-side dry run is not smart enough to resolve these dependencies when submitting a batch of resources, so this too will cause a failure. To me, this puts the multi-stage deploys of krane fundamentally at odds with the ability to get a reliably-passing batch dry-run and is the largest argument against continued support for batched dry-runs.

[jpfourny]

❤️ for splitting test matrix by test suite.