Warn against using emptyOrUnsetBlockDiffSuppress

[rileykarson]

After #6546 fixed some GKE problems, I dug into emptyOrUnsetBlockDiffSuppress a bit. I think we should recommend against using it- it's harder to understand, and is only used in two places so far.

I think what it really does is allow a user to specify an empty block when the API returns nil, which is pretty different than what O+C does. But it reads like an O+C replacement initially (and it somewhat is one). Also, I think it behaves weirdly with some values returned from the API.

If this PR is for Terraform, I acknowledge that I have:

• Searched through the issue tracker for an open issue that this either resolves or contributes to, commented on it to claim it, and written "fixes {url}" or "part of {url}" in this PR description. If there were no relevant open issues, I opened one and commented that I would like to work on it (not necessary for very small changes).
• Generated Terraform, and ran make test and make lint to ensure it passes unit and linter tests.
• Ensured that all new fields I added that can be set by a user appear in at least one example (for generated resources) or third_party test (for handwritten resources or update tests).
• Ran relevant acceptance tests (If the acceptance tests do not yet pass or you are unable to run them, please let your reviewer know).
• Read the Release Notes Guide before writing my release note below.

Release Note Template for Downstream PRs (will be copied)

[shuyama1]

Totally make sense! I think adding O+C plus some modifications to the expanders and flatteners should cover all the cases in emptyOrUnsetBlockDiffSuppress.

[modular-magician]

Hi! I'm the modular magician. Your PR generated some diffs in downstreams - here they are.

Diff report:

Terraform GA: Diff ( 1 file changed, 2 insertions(+))
Terraform Beta: Diff ( 1 file changed, 2 insertions(+))
TF Validator: Diff ( 2 files changed, 3 insertions(+), 3 deletions(-))

[rileykarson]

Yeah, specifically we'd want to return an empty object instead of nil if the API returned nil. We do that in a few places, but it wasn't how we wrote MMv1's expanders. Honestly not sure if that was for a good reason or not- we may have wanted to avoid O+C more strongly at the time or something, as nil works better when everything goes right.

[modular-magician]

The provider crashed while running the VCR tests in REPLAYING mode
Please fix it to complete your PR
View the build log