Define and implement the Ceph-backed volume lifecycle

[halfcyan]

Summary

Allow VM disks to use Ceph RBD without requiring a full local image copy on every compute node.

Scope

• Document local-file versus RBD behavior.
• Define volume ownership, attachment, mapping, teardown, and deletion semantics.
• Implement RBD volume resolution/mapping through an abstraction that can be mocked.
• Make mapping and teardown idempotent.
• Add configuration for pool/image references.
• This issue defines the storage model Odorobo will use after adoption; it does not migrate existing Odorobo deployments.

Tasks

• Document local-file, RBD, and image-copy behavior.
• Define the initial supported storage contract
• Define volume ownership, attachment, mapping, teardown, and deletion states.
• Introduce a mockable storage-provider abstraction.
• Implement Ceph RBD URI parsing and validation.
• Implement RBD mapping and block-device resolution.
• Track active attachments so shared volumes are not unmapped prematurely.
• Make mapping, unmapping, and cleanup idempotent.
• Add configuration for Ceph pools, images, and credentials.
• Add mocked storage-provider tests for success and failure paths.

What this should do

• A VM can reference an RBD volume through the Odorobo manifest/URI.
• The VM does not require a node-local full-image copy.
• Mapping failures produce actionable VM errors.
• Unmapping only occurs when the volume is no longer in use.
• Tests do not require Ceph in CI.

Related issues

• Related to QEMU Storage Daemon integration #10 (QEMU Storage Daemon integration) and VM Agents: Configuration #16 (VM agent configuration).
• Enables draft issue 06.

Issue links

• Related to existing #10 and #16.
• Enables #103.
• Uses the manifest contract from #98.