proper error message for fn eval/render when there are non-KRM resources

[natasha41575]

partially addresses #2039, improves error message for fn eval and fn render when there are non-KRM resources

Before:

[FAIL] "gcr.io/kpt-fn/set-namespace:v0.1.3"
  Results:
    [ERROR] failed to run function: error unmarshaling JSON: while decoding JSON: Object 'Kind' is missing in '{"Non-krm-resource":true,"metadata":{"annotations":{"config.k8s.io/id":"1","config.kubernetes.io/index":"0","config.kubernetes.io/path":"nonkrm.yaml"}}}'
  Stderr:
    "[error] /// : failed to run function: error unmarshaling JSON: while decoding JSON: Object 'Kind' is missing in '{\"Non-krm-resource\":true,\"metadata\":{\"annotations\":{\"config.k8s.io/id\":\"1\",\"config.kubernetes.io/index\":\"0\",\"config.kubernetes.io/path\":\"nonkrm.yaml\"}}}'"
  Exit code: 1

After:

Error: input resource list must contain only KRM resources:
non-krm.yaml: resource must have `apiVersion`

Will do the same for fn source and fn sink in a separate PR.

[droot]

I am actually thinking of another approach. It's important to enforce this invariant on input/output resources of each function execution(doing it in fnruntime can cover both eval and render). Use the path/index to derive a user-friendly location (2nd resource in <some.yaml>) to report an error (avoid dumping the yaml content). One caveat here: for generated resources, the path will be auto generated one, but that's okay I think.

Doing above is not sufficient to ensure the package is valid because if a package doesn't have any pipeline, function runner will not be invoked, we can have an explicit call for such cases (I think , in function named runPipeline, we return if pipeline.IsEmpty(), we can plug call for KRM check in that case). [This is not an issue for fn eval]

So, tests should include case where non-KRM resources with no pipeline as well.

[natasha41575]

I am actually thinking of another approach. It's important to enforce this invariant on input/output resources of each function execution(doing it in fnruntime can cover both eval and render). Use the path/index to derive a user-friendly location (2nd resource in <some.yaml>) to report an error (avoid dumping the yaml content). One caveat here: for generated resources, the path will be auto generated one, but that's okay I think.

Will do

[natasha41575]

@droot I moved the check to fnruntime and added a test case for no pipeline, thanks!

[natasha41575]

@droot please take a look, thanks

[frankfarzan]

please wait for @droot

[droot]

New approach is neat! Thank you.