Support importing into or merging with an existing dataset

[alexschmander]

Problem

Explicit ground-truth import can only create a new dataset; operators cannot safely merge supported imported content into an existing dataset.

Current behavior

Import builds/verifies a complete dataset and publishes it atomically with no replacement, which deliberately avoids complex identity, config, and audit-history merges.

Desired behavior

Define and implement a merge/import workflow that preserves no-replace safety, image identity, configuration compatibility, and reconstructable provenance.

Acceptance criteria

• Define matching/conflict policy for dataset/task/class/image/annotation identities and duplicate paths/content.
• Plan and preview all mutations before an authorized commit.
• Represent imported changes as replayable events with exact source/profile provenance and coverage semantics.
• Use a durable, resumable transaction with rollback/no partial publication.
• Cover active assignments, concurrent changes, schema/config mismatch, duplicate images, cancellation/restart, statistics, and snapshots.

Scope

Merge supported explicit ground-truth import profiles into an existing dataset; no predictions, segmentation, or remote acquisition unless separately supported.

Implementation notes

Preserve BLAKE3 image identity and never implement by copying files into managed paths.

Tracking source

docs/tracking/feature-requests.md:40

Import into or merge with an existing dataset without weakening no-replace publication and audit-history guarantees.