[FEATURE] ate-setup: stop staging micro-VM assets to GCS #3671

#2092 removed the GCS staging scripts (hack/microvm-assets/stage-to-gcs.sh, hack/install-microvm-deps.sh) by porting them into ate-setup, so core still uploads micro-VM assets to a GCS bucket on every non-kind cluster. Remove the GCS path and keep a provider-neutral asset build.

Tracks "Remove GCS micro-VM asset staging" in the Provider-Neutral Core workstream of the Extensible Install Roadmap.

Current state (0b635b80): cmd/ate-setup/internal/steps/microvm.go runs stageMicroVMAssetsToGCS on non-kind clusters, requires gcloud, and derives the bucket name <BUCKET_NAME>-assets. The same behaviour is documented in cmd/ate-setup/README.md, cmd/ate-setup/internal/cmd/deploy.go and differences.md.

Required changes (two steps)

1. Warning release: warn when the GCS staging path runs. Accept an operator-supplied asset location (for example --microvm-asset-location from ate-setup: install a provider's object store plugin with --component and storage.backend=external #3651).
2. Remove stageMicroVMAssetsToGCS, the gcloud requirement, and the bucket naming. A non-kind install with micro-VM and no asset location fails with a clear error. Update the docs listed above.

Dependencies: substrate-gke stages micro-VM assets itself (ai-on-gke/substrate-gke#88) and passes the location.

Breaking change: step 2.

Enforcement

• No gcloud or gs:// references remain in steps/microvm.go.
• A unit test asserts the missing-location error.

Parent Issue: #3161