Add canonical document and ingestion schema models with RawAsset support

[AymanL]

PR Summary (canonical document schema)

This PR introduces the canonical document schema foundation, aligning existing ingestion objects with a more explicit “canonical document” (definite source of truth) model vocabulary.

What changed

• Added RawAsset abstraction (eu_fact_force/ingestion/models.py)
  • Introduced abstract RawAsset(TimeStampedModel) with shared raw-asset fields/behavior.
  • Updated SourceFile to inherit from RawAsset while preserving existing fields/relations and keeping delete_source_document_from_s3() as a backward-compatible wrapper.
• New canonical models (eu_fact_force/ingestion/models.py)
  • Document: stores provider-agnostic metadata (doi, title, authors_raw, published_date, source, language, document_type, external_ids) plus optional links to SourceFile (raw_asset) and the primary ParsedArtifact.
  • IngestionRun: records ingestion attempts with status/error/provider and links to Document and/or SourceFile.
  • ParsedArtifact: represents parsed outputs tied to a SourceFile, optionally linked to Document, including format, storage_key, page_count, parser info, and stats.
• Extended DocumentChunk (eu_fact_force/ingestion/models.py)
  • Added canonical chunk metadata fields: document, parsed_artifact, chunk_type, page_number, start_offset, end_offset.
  • Kept existing chunk/search fields (content, order, embedding) unchanged for compatibility.
• Updated Django admin (eu_fact_force/ingestion/admin.py)
  • Registered the new models (Document, ParsedArtifact, IngestionRun) and expanded DocumentChunk admin display to include the new fields.
• Database migration (eu_fact_force/ingestion/migrations/0003_canonical_document_schema.py)
  • Creates tables for Document, IngestionRun, ParsedArtifact.
  • Adds the new fields/relations and indexing/constraints to Document and DocumentChunk.

Tests

• Added tests/ingestion/test_canonical_document_schema.py
  • Covers Document DOI uniqueness behavior (non-empty DOIs enforced).
  • Validates linkage wiring for IngestionRun ↔ Document/SourceFile, and ParsedArtifact ↔ Document/SourceFile.
  • Confirms persistence of new DocumentChunk metadata fields.
• Verified by running: uv run pytest tests/ingestion/ -v (20 tests passed).