Task: Codegen per-entity history field mapper

Add a new presentation codegen facet: render_{entity}_fields(), a plain, generated (not hand-written, not runtime-reflective) per-entity function converting a domain value to an ordered ores::diff::domain::field_value list, per History Diff Architecture layer 1.

• New presentation facet (ores.cpp.presentation, both the literate doc group and the physical-space address node) declares/implements render_{entity}_fields(const domain::{entity}&) per entity.
• Generated for the same 27 ores.refdata entities as entity_type_of() (task 352B308F) — the established clean baseline for this story; the same 7 drifted entities stay excluded.
• One field_value push per field: primary key, natural keys, columns, audit columns, recorded_at. Labels are mechanical title-case, not hand-curated.
• Handles every cpp_type actually present across the 27 entities: std::string, std::optional<std::string>, int, std::optional<int>, double, std::optional<double>, bool, std::optional<bool>, boost::uuids::uuid, std::optional<boost::uuids::uuid>, plus recorded_at's fixed std::chrono::system_clock::time_point via the existing ores::platform::time::datetime::to_iso8601_utc().
• Local build clean (full repo, -j3); full ctest suite green (69/69 suites, 215s).

1. Investigated whether the existing per-column type flags (is_uuid=/=is_nullable_string=/=is_already_optional=/ =is_plain_string, used by the repository/SQL facets) could be reused directly for the mapper. They could not — see Notes for why; this cost a full round of regenerate → compile-fail → diagnose → fix.
2. Added a new, independent set of render_is_* column flags in core.py, derived directly from each column's raw cpp_type string (the same ground truth the domain class template itself emits verbatim) — render_is_string, render_is_optional_string, render_is_bool, render_is_optional_bool, render_is_int, render_is_optional_int, render_is_double, render_is_optional_double, render_is_uuid, render_is_optional_uuid, render_is_timestamp, render_is_optional_timestamp. Also added render_label (mechanical title-case) to columns, natural_keys, and primary_key — none had a label before (the closest existing thing, the Qt Detail Fields table's curated label column, is about widget wiring and isn't guaranteed present for every entity).
3. Added the ores.cpp.presentation facet: two archetypes (cpp_history_field_mapper.hpp/cpp.mustache, tangled from ores.cpp.presentation.history_field_mapper_header/impl.org), a facet-group literate doc (cpp_presentation.org), a facet_catalogue.org entry, and the separate physical-space address node (ores.cpp.presentation.org) the --address resolution mechanism requires (distinct from the literate group doc — see Notes).
4. Proved the facet on currency alone first via codegen generate --model ... --profile ores.cpp.presentation (note: --profile <name> only works when <name> is the real ores.cpp.<facet> address, not a bare facet-catalogue heading — see Notes).
5. First attempt used the wrong flags (repository/SQL layer's nullable-promotion flags) and produced code that failed to compile against book (uuid primary key rendered as a plain string; nullable text columns with an explicit non-optional cpp_type override called .value_or() on a plain std::string). Root- caused (see Notes) and replaced with the render_is_* flags above.
6. Regenerated all 27 clean entities via codegen regenerate --component refdata-cpp --profile ores.cpp.presentation; removed the 7 known-drifted entities' output to stay consistent with the established baseline.
7. Added ores.diff.lib to ores.refdata.core's link libraries (public, since field_value.hpp is now in a public header).
8. Added projects/ores.refdata/core/tests/presentation_currency_history_field_mapper_tests.cpp covering field count/uniqueness, string/int rendering, optional- uuid (both empty and populated), and recorded_at formatting.
9. Verified: full local build clean, full ctest suite green (69/69).

Why the repository/SQL layer's type flags don't apply to the domain class. is_uuid=/=is_nullable_string=/=is_already_optional=/ =is_plain_string encode a nullable-to-std::optional-promotion decision that only the repository entity and SQL facets apply. The domain class template (cpp_domain_type_class.hpp.mustache) does not apply that promotion — it emits \{\{\{cpp_type\}\}\} verbatim. A column can have :nullable: true in its model but an explicit :cpp_type: std::string override, and the domain struct gets a plain std::string field, not std::optional<std::string> — exactly what happened with book's description=/=gl_account_ref=/=cost_center, and what made the first implementation attempt fail to compile. The render_is_* flags fix this by deriving straight from the raw cpp_type string — the same source of truth the domain class template itself uses — rather than from the derived nullable-promotion flags.

Two separate facet-registration files are needed for a new facet: the literate group doc (e.g. cpp_presentation.org, human-facing, indexes archetypes) is not the same thing generator.py's --address resolution reads. That reads a physical-space address graph built from files matching ores.*.org in the templates directory (ores.cpp.presentation.org here), each declaring #+type: facet, #+model_types:, and #+facet_group:. Forgetting this file produces "unknown address" errors that look like a typo but aren't. --profile <name> only works as a bare facet name for the small set of legacy curated profiles (domain, sql, all-cpp, …); any other name is treated as a literal address and must be the full ores.cpp.<facet> form.

Regenerating a new facet is comparatively low-risk for drift (no prior generated files exist to diff against — every output is a new file), unlike modifying an existing facet's template. The risk here is instead per-entity field-type coverage gaps, not silent drift.

Scope not covered by this facet (deferred, matching precedent from entity_type_of and the redesign task):

• has_identity_group=/=has_audit_group entities (none of the 27 covered entities use them).
• Enum columns (is_enum; none present in the 27 covered entities).
• The 7 entities excluded from entity_type_of() for pre-existing model drift (business_unit, contact_type, currency_market_tier, monetary_nature, party, party_id_scheme, rounding_type) — party in particular is a known gap the generic-history-subject task inherits.
• Junction models (not in the domain=/=presentation facets' model_types scope at all).

Manual QA scenarios (scaffolded via compass add test_scenario, run through the QA Validation Runner panel) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

Added the ores.cpp.presentation codegen facet: render_{entity}_fields() per entity, generated for all 27 clean ores.refdata entities (same baseline as entity_type_of()). First implementation attempt reused the repository/SQL layer's nullable-promotion type flags and produced code that failed to compile against real entities (book's uuid primary key, and nullable-but-explicitly-non-optional string columns) — root-caused to the domain class template using raw cpp_type verbatim rather than a promoted type, and fixed with a new, independent render_is_* flag set derived from that same raw cpp_type ground truth. Added ores.diff.lib as a public dependency of ores.refdata.core. New test coverage in presentation_currency_history_field_mapper_tests.cpp. Full local build clean; full ctest suite green (69/69 suites, 215s).

Distilled into the story's Decisions: field-mapper rendering logic must derive from raw cpp_type, never from the repository/SQL layer's nullable-promotion flags — the two facets diverge whenever a column has an explicit cpp_type override.