Story: Convert codegen mustache templates to literate org-mode documents

Every codegen mustache template is owned by a literate org-mode document. Today projects/ores.codegen/library/templates/ holds 116 raw .mustache files with no narrative: nothing explains what model fields a template consumes, what artefact it produces, or why its structure is the way it is. The project already has a working literate pattern — doc/knowledge/architecture/clang_format_config.org tangles to .clang-format, doc/llm/claude_code_settings.org tangles to .claude/settings.json — and this story extends it to the codegen template library.

After this story, the org document is the single source of truth for each template: it documents the template's role, its model inputs and its output artefact, and tangles to the .mustache file consumed by codegen. The library is organised as a three-level literate hierarchy — template ← facet doc ← group doc ← groups overview: each facet group gets a "namespace" org file documenting the group, and a top-level file describes all the groups. The tangled files carry a generated-file header naming their source document, and a zero-diff tangle check makes drift between org and mustache detectable.

Promoted from inbox capture 76178191-E20D-4479-BC01-79B877247D03, which lives on as the rollout task below.

• All templates under projects/ores.codegen/library/templates/ have a corresponding literate org document that tangles to them.
• Tangled .mustache files carry a generated-file header naming the source org document.
• Tangling is reproducible via the project tangle tooling — no manual steps.
• Codegen output is byte-identical before and after the migration (zero-diff regeneration).
• A drift check (zero-diff on tangle) exists so CI or compass can detect divergence between org sources and tangled templates.

• Granularity: one org file per facet, with each template a #+begin_src mustache :tangle <name>.mustache block preceded by prose (role, model inputs, output artefact, profiles). Groups (cmake, cpp, sql, doc, assets) get <slug>_group.org namespace docs; overview.org tops the hierarchy.
• Facet cut: facets are cohesive MASD facets — entity-lifecycle layer × profile membership — not filename-prefix buckets, and stay at roughly a dozen templates or fewer without rationale. The cpp re-cut (PR #1136) set the pattern: cpp_domain, cpp_table, cpp_repository, cpp_qt, cpp_messaging, cpp_service_app, cpp_component, cpp_model_types. IDs stay stable when a facet merely renames; moved blocks get their GENERATED FILE header repointed (render-neutral, header-only artefact diffs).
• Location: org sources co-located with their artefacts in projects/ores.codegen/library/templates/.
• Tangle tooling: ores-build-codegen-templates.el behind the tangle_codegen_templates CMake target; tangles every org doc in the templates directory (group docs/overview are no-ops).
• Header: every tangled file starts with {{! GENERATED FILE — tangled from <facet>.org. Edit the org source. }} — standalone mustache comments are spec-stripped, so pystache output is byte-identical (proven on all eight cmake templates).
• Drift checks: tangle then git diff --exit-code -- projects/ores.codegen/library/templates for org↔mustache drift; regenerate_facet_inventories.py --check for group-inventory drift. CI wiring lands with the rollout task.
• Doc types: compass add facet (requires --facet-group, lands #+facet_group: for the inventory generator) and compass add facet_group scaffold the hierarchy; facet inventories in group docs are generated, never hand-edited.
• Link rule: repo files not published as HTML (e.g. profiles.json) are linked with proj: (GitHub blob URLs), not file: — repo-wide cleanup tracked by the replace-file-links task.

• Changing template content or codegen behaviour — this is a format/documentation migration; generated output must not change.
• Migrating the entity models to org — that is the in-flight Codegen unified model — org-mode migration story.