chore!: document the schema-first workflow, adopt gen3schemadev 3.1.0, remove dead scaffolding

[JoshuaHarris391]

This repository's model is maintained by hand — the node YAMLs in dictionary/prod_dict/ are the
source of truth. Nothing said so. The README documented a "generate into test_dict, copy elements
into prod_dict" loop that nobody follows, so a new contributor had no way to know where the model
actually lives.

Why input_dd.yaml had to go

It was a fork artefact, not a build input — no property definitions at all, just 27 node stubs
and 26 link triples, frozen since 2025-11-10 while the dictionary moved on across 29 commits.

Regenerating from it would have:

Properties destroyed	343 of 448 (77%) — proteomics_assay 37 → 2, medical_history 31 → 2
Custom enum definitions wiped	all of them, leaving dangling $refs across the node files
Links dropped	11 of 45, including every extra parent on qc_file, supplementary_file, analysis_workflow
project.yaml / program.yaml	replaced by a generic template, losing systemProperties and uniqueKeys — breaks Gen3 submission
_dict_version	regressed 1.2.0 → 0.5.0

The README now states the rule and the reason, rather than just forbidding it.

Removed

dictionary/input_dd.yaml, dictionary/test_dict/ (a byte-identical mirror of prod_dict), both
copy scripts, scripts/add_enum_to_def.py with its stale CSV, and notebook.ipynb.

On the enum script: it existed to re-inject enums after generate overwrote _definitions.yaml, and
nothing generates. Nothing referenced it. _definitions.yaml is now the richer record anyway — 373
enumDef entries against the CSV's 195 rows, with more source and term_id annotations — so
deleting the CSV loses no provenance.

gen3schemadev 2.6.1 → 3.1.0

For a validate-and-bundle-only repository the value is specific: generate now refuses to
overwrite existing files, so the accidental wipe described above is no longer possible. The bundle
is re-serialised in sorted key order as a side effect; content unchanged.

CI extended

It previously validated only the loose YAMLs. Now also:

• the bundle is validated in its own right — it is what Gen3 loads, and was never checked
• bundle freshness — a branch can no longer edit a schema and forget to re-bundle
• version consistency between _settings.yaml and pyproject.toml

Each gate was negative-tested: editing a schema without re-bundling, mismatching the version, and
adding a property with no type all fail as they should.

Verification

No node schema is modified. All 32 dictionary YAMLs are byte-identical to main, verified by
hashing before and after — the check that proves nothing regenerated. The bundle's content is
identical to what was committed, differing only in key order, confirmed by parsing both. All four
gates pass locally, bundling twice is byte-stable, and every README link resolves.

PR #8 and its branch are untouched.

Follow-ups, not done here

• pyproject.toml is brought to 1.2.0 to match _dict_version, but no tag is cut — releasing is
  a separate decision and six environments pin versions independently
• docs/acdc_proteomics_metabolomics_methods_and_rationale.md references a companion implementation
  spec that does not exist in the repository
• Dependabot reports 19 vulnerabilities on the default branch, unrelated to this change

🤖 Generated with Claude Code