Expand Relational Workflow Model concept page

[dimitri-yatsenko]

Context

The current Relational Workflow Model (RWM) concept page (src/explanation/relational-workflow-model.md) is understated relative to the model's significance. It reads as a brief positioning statement rather than as an entry point that lands the structural argument for a reader who already knows informatics (databases, FK graphs, ER modeling, workflow managers, lakehouses).

This PR expands the page to function as that entry point — the audience pictured is a knowledgeable peer (e.g., an infrastructure architect from pharma R&D evaluating where DataJoint sits in the landscape they already know).

Changes

• Lead with the three-interpretations taxonomy (Codd / Chen / RWM) and the computational substrate framing from the DataJoint 2.0 preprint (Yatsenko & Nguyen, 2026, arXiv:2602.16585).
• Name the surrounding tool categories explicitly and what each is silent on:
  • File-based workflow systems (CWL, Snakemake, Nextflow) — fragment provenance across the filesystem
  • Task orchestrators (Airflow, Argo, Prefect) — agnostic to data structure
  • Data catalogs (DataHub, Atlan, Marquez) — describe data after it lands
  • Lakehouses (Delta, Iceberg, Hudi) — treat computation as external
• Add a worked-example pipeline (Mouse → Session → Scan → AverageFrame → Segmentation → Fluorescence, with SegmentationParam as Lookup) rendered as a mermaid diagram with tier-color classes.
• Add a "deliberate trade-off" section that acknowledges the legitimate strengths of decoupled architectures and frames DataJoint's coupling as a chosen trade-off — directly drawn from the preprint Section 5.
• Add a "substrate consequences" section that covers:
  • Provenance and lineage as structural properties of the substrate (mapping to W3C PROV / OpenLineage is translation, not reconstruction)
  • The five agent-substrate properties from the preprint: self-describing, safe by default, explicit dependencies, idempotent, observable
• Preserve the existing detailed sections (table tiers, master-part, workflow normalization, entity integrity, query algebra with closure, transactions vs transformations) under a "Beneath the model" header for readers who want the structural detail.

Net change

+177 / -112 lines; one file.

Sources

• Yatsenko & Nguyen, 2026 — DataJoint 2.0 whitepaper (computational substrate, four innovations, substrate properties for agents, deliberate-trade-off discussion)
• Yatsenko et al., 2018 — original theoretical formalization (relational workflow model, query algebra)

Notes for reviewers

• The mermaid diagram uses tier-color classDefs. If the docs site's mermaid theme overrides these, we may need to drop colors or adapt to the site theme.
• Cross-references in See also all resolve against current src/explanation/ and src/how-to/ content.