CRDT tests

Deterministic scenarios

Small named tests describe one user-visible operation history at a time.

Entity	Operations and behavior covered
Media	Creation, rename, property updates, delayed creation delivery, and same-counter rename conflicts.
Album	Creation, rename, reparenting, thumbnails, media add/remove, deletion, missing parents, deleted ancestors, and cycles.
Group	Creation below an album, media add/remove, concurrent membership changes, deletion, and deleted-parent visibility.
State	Mixed media/album/group histories, duplicate delivery, Lamport-clock advancement, and persisted snapshots.

Each scenario checks:

The expected library result: which media, albums, groups, and memberships are visible.
Every delivery order for its small operation list. The operations are applied in that order, then the resulting CrdtState must converge to the same exact state.

fuzzy_multi_devices_convergence tests longer histories that are too large to permute. It runs 64 seeded simulations.

Each simulation:

Uses 2–5 in-memory devices over 24–48 rounds.
Includes one shared in-memory remote operation store.
Each device records the index of the last remote entry it fetched.
Ends with one final convergence round: every device pushes to the remote, then fetches the remote's complete contents.

With 0.90 probability, run an action round:

Select a random device.
With 0.35 probability, fetch every remote-log entry after the device's recorded index.
Draw and apply 1–10 operations valid for the selected device's current state.
Choose operation weights once from the simulation seed:
- Media creation has a weight from 1 to 3; media rename and property updates have weight 1.
- Album creation, album deletion, group creation, and group deletion form two pairs. Each deletion has weight 1 or 2; its creation weight is 1 or 2 higher.
- Album-media and group-media membership form two pairs. Each removal has weight 1 or 2; its add weight is 1 or 2 higher.
- Album rename, reparenting, and thumbnail updates have weight 1.
With 0.40 probability, push every operation the remote does not yet know.
A delivered batch has a 0.10 probability of being delivered twice.
Check that the selected device matches the state obtained by merging the operations it knows.

With 0.10 probability, run a convergence round:

Every device pushes its known operations to the remote.
Every device fetches every remote-log entry after its recorded index.
Check that every device's CrdtState equals the state reconstructed from the remote operations, except for its local device_id.

Run CRDT tests with:

cargo nextest run -p lasco-core crdt::