Support orphaned resources cleanup for Nebius - #4412

Follow-up to #4401 for #3453. Nebius provisioning regularly leaks boot disks and pricing policies: the rollback in create_instance() waits 25s for the instance to be deleted while deletion takes ~100s, and a pricing policy that is still being created can't be deleted.

• Support orphan cleanup for Nebius: the instance, boot disk, and pricing policy of a create_instance() call share a name with provider_ref encoded, and the disk and instance are also labeled with dstack_provider_ref and dstack_provider_name, since Nebius drops labels on pricing policies.
• List each region's project and group resources by that name into one group per provisioning attempt; an instance being deleted is not listed as the owner, but its id is kept so that its disk and pricing policy are deleted after it's gone.
• Share the deletion with terminate_instance(): the instance first, then the boot disk and pricing policy, raising NotYetTerminated while a pricing policy is still being created instead of failing.
• Fix an UnboundLocalError in the provisioning rollback when the instance is already gone.
• List nebius among the backends supporting orphan cleanup.