Velero data movement - generic volume restore design

[Lyndon-Li]

We want to create a design for a generic solution to restore volumes for Velero data movement. Basic workflow is as follows:

• Create the workload pods
• Provision PVs to the pods through the generic kubernetes approach, for example, dynamic provision or static provision
• On hold the workload pods
• Expose the PVs to the data mover pod (through host path vs. attach to pod)
• Inside the data mover pod, call the related uploader to copy data from the backup repository to the PV
• Release the workload pods

At present, two candidates:

• Velero's own approach that is similar to pod volume restore
• Kuberntes' built in Volume Populator mechanism

Benefits of building this generic restore solution:

• The solution works for all kinds of backups, i.e., CSI backups, pod volume backups, direct access backups and any other backups added in future
• The solution doesn't rely on any underlying storage snapshot mechanisms, for example, the restore process doesn't require the CSI snapshot objects or workflows. Therefore, it supports to restore to a heterogenous environment
• For the same reason, it doesn't need to maintain the underlying storage snapshot objects and so leave many snapshot objects to the production environment after restore

[reasonerjt]

This will be covered in v1.11 when we deliver the overall design of data mover in velero.
Closing as dup of #4112