feat(volumes): btrfs storage pools #339

Problem

Docker local volumes have no snapshots, no quotas and no cheap copies. Consistent backups of a busy database need downtime (stop) or application tooling (dump); one volume can fill the disk; and a copy of a volume costs a full copy.

Proposed solution

Storage pools, declared in config.toml because they are facts about the host (#334):

[[storage.pools]]
name = "data"
kind = "btrfs"
path = "/srv/piqueld"

local, today's Docker volumes, stays the default and needs no declaration. A btrfs pool creates one subvolume per volume and bind-mounts it through a local volume. It needs only a btrfs filesystem at path, such as a second disk or an attached block volume, not a btrfs host.

• Atomic snapshots: the snapshot backup method (feat(backups): scheduled environment backups, with their volumes #332) copies a read-only snapshot, with no downtime.
• Quotas: a volume's size in the manifest is enforced with btrfs simple quotas (kernel 6.7 or newer), and reported as a warning on local pools.
• Clones: a btrfs snapshot is a writable, independent subvolume, so a volume can be created from another in seconds, and deleting the original never affects the copy. feat(previews): seed preview data from branches or backups (idea) #333 builds on this.
• The default pool is a daemon setting; piquelctl volume move VOLUME --pool moves a volume through a stopped copy.

ZFS later, as a second kind behind the same interface, for hosts that already run it. Its refquota and zfs send matter more once nodes replicate volumes (#325), but its clones depend on their origin snapshot and it is an out-of-tree module.

The NixOS module gains the pool declarations.

Written with Claude Opus 5.5 in Claude Code (via T3 Code).