fix(gcsartifact): prevent artifact version race in concurrent Save

[wolo-lab]

Problem

gcsService.Save assigned version numbers non-atomically: it listed the
existing versions, computed max+1, and wrote the blob at that version. Two
concurrent Save calls for the same artifact read the same version set, pick
the same next version, and the second upload silently overwrites the first —
callers get duplicate version numbers and one object is lost. The code carried a
TODO acknowledging the race.

Note: adk-python's GcsArtifactService has the same unguarded
max(versions) + 1 logic, so this is Go-specific hardening rather than a port.

Summary

• Write the new blob with a "does not exist" generation precondition
  (x-goog-if-generation-match: 0) via
  ObjectHandle.If(storage.Conditions{DoesNotExist: true}). GCS has no
  transaction spanning the list and the write, so a colliding writer now fails
  the precondition with HTTP 412 instead of clobbering the existing object.
• On a 412, Save re-reads the versions and retries with a fresh number,
  bounded by maxSaveAttempts (returns errVersionConflict if exhausted). N
  concurrent saves therefore yield exactly versions 1..N with no lost writes.
• Extracted writeArtifact / isPreconditionFailed helpers and added an
  ifNotExist() method to the internal gcsObject seam and its wrapper. No
  public API change