nvme_driver: Set correct drain state for new queues (#2864)

[alandau]

Clean cherry pick of PR #2864

In #2833, saved queues (eager and proto) were given the correct drain state, but new queues were given the AllDrained state meaning they could race before all existing queues have drained. This could happen if IO is received during draining on a CPU (thus queue) that has never seen IO before the save.

This PR fixes the race by giving new queues the correct drain state - either AllDrained if all existing queues have drained already, or SelfDrained, like proto queues (no IO pending on them) if not.

[Copilot]

Pull request overview

Fixes a restore-time race in the NVMe driver’s “drain-after-restore” barrier by ensuring newly created I/O queues (created after restore, not from saved state) receive an appropriate drain state instead of incorrectly starting in AllDrained.

Changes:

• Add DrainAfterRestoreBuilder::{is_drain_complete, new_for_new_queue} to pick the correct drain state for newly created queues during an active drain.
• Thread an explicit DrainAfterRestore into QueuePair::new so new queues don’t implicitly default to AllDrained.
• Store a shared DrainAfterRestoreBuilder in the worker during restore so new queues participate in the drain barrier until completion.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
vm/devices/storage/disk_nvme/nvme_driver/src/queue_pair.rs	Adds builder helpers to determine drain completion and compute the correct drain state for brand-new queues; updates QueuePair::new to accept an explicit drain state.
vm/devices/storage/disk_nvme/nvme_driver/src/driver.rs	Persists the drain builder across restore so newly created queues can block on draining; uses new_for_new_queue() when creating new I/O queues.