kernel: ingest pipeline — SHA-256 hashing, blob-store dedup, technical metadata extraction, materialization into a Batch

[JArmandoAnaya]

Context

The pipeline that IngestJob executes, gluing MediaProcessor + BlobStore + metadata.

Deliverables

• Per item: decode/validate → SHA-256 → BlobStore.put (skip if exists) → Asset row with
  modality=image, dimensions, format, content_hash, origin (source + frame index/timestamp)
  → membership in the target draft Batch.
• Asset identity = content hash: the same image in two sources yields one blob and
  hash-stable asset identity (document the Asset-vs-blob identity decision in the model
  docstrings).
• Batch is created in draft by the ingest (or an existing draft batch is targeted).

Acceptance criteria

• Duplicate image across two sources ⇒ one blob
• Asset origin metadata correct for both directory and video paths
• Pipeline covered end-to-end with generated fixtures