Prune read-only fields from resource inputs

[EronWright]

Proposed changes

This PR improves the handling of the read-only metadata fields of Kubernetes objects, such as creationTimestamp and resourceVersion. The provider now ignores these fields when they appear as resource inputs, to address a few quirks:

1. Unparseable values do not cause an API Server error during preview.
2. A subsequent refresh does not show a difference to the __inputs property.
3. The behavior is now uniform for standard resource types and for custom resource types.

An integration test is provided to show that such fields are ignored as inputs, and are still available as outputs. Manual tests confirmed identical behavior for custom resource objects.

The resultant state shows that the read-only fields are not present in "inputs" nor in "__inputs", and are present in "outputs":

{
                "urn": "urn:pulumi:dev::issue-2351-simple-ts::kubernetes:apps/v1:Deployment::nginx",
                "custom": true,
                "id": "default/nginx-78a03128",
                "type": "kubernetes:apps/v1:Deployment",
                "inputs": {
                    "apiVersion": "apps/v1",
                    "kind": "Deployment",
                    "metadata": {
                        "annotations": {
                            "pulumi.com/autonamed": "true"
                        },
                        "name": "nginx-78a03128",
                        "namespace": "default"
                    },
                },
                "outputs": {
                    "__fieldManager": "pulumi-kubernetes-42ad785a",
                    "__initialApiVersion": "apps/v1",
                    "__inputs": {
                        "apiVersion": "apps/v1",
                        "kind": "Deployment",
                        "metadata": {
                            "annotations": {
                                "pulumi.com/autonamed": "true"
                            },
                            "name": "nginx-78a03128",
                            "namespace": "default"
                        },
                    },
                    "apiVersion": "apps/v1",
                    "kind": "Deployment",
                    "metadata": {
                        "annotations": {
                            "deployment.kubernetes.io/revision": "1",
                            "pulumi.com/autonamed": "true"
                        },
                        "creationTimestamp": "2023-09-19T23:58:41Z",
                        "generation": 1,
                        "name": "nginx-78a03128",
                        "namespace": "default",
                        "resourceVersion": "559181",
                        "uid": "ec381c4d-92a7-470a-8fdc-824cd22a59dc"
                    },
                },
            }

Related issues (optional)

Closes #2351

Related:

• Fix continuous diff for Secret stringData field #2511
• Overhaul logic for resource diffing #2445

[github-actions]

Does the PR have any schema changes?

Looking good! No breaking changes found.
No new resources/functions.

[codecov-commenter]

Codecov Report

Merging #2571 (69cb28d) into master (d2a3b8c) will decrease coverage by 0.04%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##           master    #2571      +/-   ##
==========================================
- Coverage   18.64%   18.61%   -0.04%     
==========================================
  Files          47       47              
  Lines        9434     9440       +6     
==========================================
- Hits         1759     1757       -2     
- Misses       7576     7584       +8     
  Partials       99       99              

Files Changed	Coverage Δ
provider/pkg/clients/unstructured.go	59.04% <ø> (-0.77%)	⬇️
provider/pkg/provider/provider.go	7.37% <0.00%> (-0.03%)	⬇️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[EronWright]

I moved the pruning logic to normalizeInputs in provider.go for two reasons:

1. Pruning of read-only fields makes sense only for inputs, not for outputs, but the clients.Normalize function seems general-purpose in nature. That is, general-purpose normalization is independent of input pruning.
2. It is desirable to prune the read-only fields even for custom resources, where a scheme isn't available (and shouldNormalize would be false).

I renamed normalize to normalizeInputs in provider.go for clarity, and double-checked that it is not used with outputs.

[lblackstone]

Changes look good to me, but I'll defer to Ramon for a closer look.

[rquitales]

It looks like auto-merge didn't block until an approval was given. Regardless, the new changes LGTM.