feat: add Google Cloud Storage backend

[dylayed]

Summary

This adds Google Cloud Storage as an alternative backend on top of #2. An exact gs://BUCKET target selects GCS; existing bare bucket names and s3://BUCKET continue to select S3.

GCS uses Application Default Credentials and object generations for conditional writes. Deployment uploads, runtime bucket access, ownership, leases, and LTX replication all use the selected backend.

Approach

• Validate and normalize the storage target before control-plane enrollment. S3 keeps its existing enrollment form; GCS uses gs://BUCKET with no endpoint or region identity.
• Build GCS with the standard ADC file override, local ADC, or metadata identity. Custom GCS endpoints and managed GCP credentials are not included.
• Map GCS generations to the same opaque version value used by the ownership code. Conditional-write retries remain disabled.
• Run the existing celld-ltx object-store conformance suite against GCS beside the MinIO integration test.

This is intentionally a narrow first integration: no bucket prefixes, emulator support, or cross-provider migration.

Review

The main things to review are provider selection and enrollment normalization, ADC setup, and generation-based CAS in crates/celld/storage_backend.rs. Most other changes make existing output and documentation name the selected provider correctly.

Review #2 first. The two PRs are linked in native stack #3.

Validation

• Celld checks and the celld-ltx S3/GCS test build pass.
• Focused provider, enrollment, and ownership tests pass.
• Real MinIO regression coverage passes the CAS contract and all 3 LTX integration tests, including multipart.
• Live GCS conformance completed without a skip and cleanup left no objects behind.
• A two-worker GCS deployment preserved counter writes and restored state across worker replacement.