fix(gcsartifact): recognize wrapped storage.ErrObjectNotExist

[shoemoney]

Link to Issue or Description of Change

2. Or, if no issue exists, describe the change:

Problem:

artifact/gcsartifact/service.go has two identical checks (Load and GetArtifactVersion) that compare the error from blob.attrs(ctx) against storage.ErrObjectNotExist with ==:

attrs, err := blob.attrs(ctx)
if err != nil {
    if err == storage.ErrObjectNotExist {
        return nil, fmt.Errorf("artifact '%s' not found: %w", blobName, fs.ErrNotExist)
    }
    return nil, fmt.Errorf("could not get blob attributes: %w", err)
}

The pinned cloud.google.com/go/storage v1.64.0 always wraps this sentinel. ObjectHandle.Attrs routes through httpStorageClient.GetObject → formatObjectErr:

func formatObjectErr(err error) error {
    var e *googleapi.Error
    if s, ok := status.FromError(err); (ok && s.Code() == codes.NotFound) ||
        (errors.As(err, &e) && e.Code == http.StatusNotFound) {
        return fmt.Errorf("%w: %w", ErrObjectNotExist, err)
    }
    return err
}

So in production err == storage.ErrObjectNotExist is always false, and the fs.ErrNotExist translation is dead code. A caller doing errors.Is(err, fs.ErrNotExist) after a real not-found gets a generic "could not get blob attributes" error instead of a not-found signal.

This package already documents the fix for this exact class of bug two functions up, on ErrVersionConflict:

// ErrVersionConflict ... It is always wrapped, so test for it with [errors.Is]

This PR applies that same rule to storage.ErrObjectNotExist at both call sites.

Solution:

Change both err == storage.ErrObjectNotExist checks to errors.Is(err, storage.ErrObjectNotExist). Both errors and storage were already imported. No other behavior changes — one concern only.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Added TestNotFoundIsWrappedSentinel in artifact/gcsartifact/gcs_test.go, covering both Load and GetArtifactVersion. It adds a small attrsErr hook to the existing in-memory fake (fakeBucket/fakeObject) so attrs() can return a wrapped storage.ErrObjectNotExist — mirroring what the real client actually returns — instead of the fake's previous bare-sentinel behavior, which never exercised this bug.

Confirmed RED on the pre-fix code (test added, fix withheld):

=== RUN   TestNotFoundIsWrappedSentinel/Load
    gcs_test.go:246: err = could not get blob attributes: storage: object doesn't exist: googleapi: Error 404: Not Found, want errors.Is(err, fs.ErrNotExist) = true
--- FAIL: TestNotFoundIsWrappedSentinel/Load (0.00s)
=== RUN   TestNotFoundIsWrappedSentinel/GetArtifactVersion
    gcs_test.go:246: err = could not get blob attributes: storage: object doesn't exist: googleapi: Error 404: Not Found, want errors.Is(err, fs.ErrNotExist) = true
--- FAIL: TestNotFoundIsWrappedSentinel/GetArtifactVersion (0.00s)
FAIL

GREEN after the fix, along with the full existing package suite:

go test -race -mod=readonly -count=1 -shuffle=on ./artifact/... ./artifact/gcsartifact/...
ok  	google.golang.org/adk/v2/artifact	1.028s
ok  	google.golang.org/adk/v2/artifact/gcsartifact	1.043s

Also ran, all clean:

• golangci-lint run ./artifact/... — 0 issues
• golangci-lint fmt ./artifact/... — no changes
• go mod tidy -diff — empty
• go build -mod=readonly work — succeeds

Manual End-to-End (E2E) Tests:

Not applicable — this is a pure error-classification fix inside gcsService.Load/GetArtifactVersion, fully covered by the unit test above; no UI, runner, or live-GCS surface is touched.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end. (not applicable — see above)
• Any dependent changes have been merged and published in downstream modules.