roachprod: centralized api - full cluster lifecycle management

Problem: Currently, users must have cloud credentials locally to create, modify, or destroy clusters. This limits who can perform these operations and makes it difficult to enforce policies, track changes, or implement approval workflows. The centralized service can see cluster state but cannot manage it.

Solution: Extend the API to support full CRUD operations for clusters:
- Create clusters by specifying cloud provider, machine types, cluster size, and configuration
- Update cluster properties (resize, modify settings)
- Destroy clusters when no longer needed
- Execute these operations via a distributed task system for reliability

Operations should respect user permissions (scoped by cloud provider and account), queue as background tasks with progress tracking, and maintain audit logs.
This enables centralized enforcement of resource policies while removing the need for widespread cloud credential distribution.

Jira issue: CRDB-56181

Epic CRDB-49123