Task: Migrate fx_convention consumers to currency_pair and retire it

Point every real consumer of ores.refdata.fx_convention — ores.ore's conventions_mapper::map_fx=/=reverse_fx, the Qt CRUD screens (FxConventionController=/=DetailDialog=/=MdiWindow=/etc.), and the CLI registration in =ores.cli/src/app/application.cpp — at currency_pair instead, per the field mapping settled in Reconcile currency_pair design with fx_convention. Then delete fx_convention's SQL table/triggers, repository, service, messaging, Qt files, and generator — it has no other reason to exist once nothing references it.

Depends on the currency_pair ores.codegen.entity implementation existing first (not yet a task in this story — needs scaffolding).

• [X] ores.ore's conventions_mapper reads/writes currency_pair (via its points_factor() helper for the divisor form) instead of fx_convention; the exported/imported ORE XML is unchanged byte-for-byte for the fields that carry over.
• [X] fx_convention's Qt CRUD screens and CLI registration are removed, not just left dangling; any equivalent CRUD need is served by currency_pair's own generated Qt/CLI surface.
• [X] fx_convention's SQL table, triggers, repository, service, and messaging are deleted from the codebase, not merely deprecated.
• [X] No dangling references to fx_convention remain (grep clean).

1. Rewrote conventions_mapper::map_fx=/=reverse_fx (ores.ore.core) to map fxType to/from a new mapped_fx transport struct pairing currency_pair + currency_pair_convention. SpotDays is carried on mapped_fx itself (not persisted on either domain type) because the ore roundtrip import/export path has no database access and must still round-trip the value byte-for-byte — see the currency-derived spot-lag note below for why it can't be looked up from currency here. Verified byte-for-byte roundtrip against external/ore/examples/Input/conventions.xml's <FX> elements.
2. Repointed ores.cli's import_conventions at currency_pair_repository=/=currency_pair_convention_repository instead of fx_convention_repository.
3. Repointed the Qt "ORE Conventions" menu (RefdataPlugin) at CurrencyPairController=/=CurrencyPairConventionController instead of FxConventionController — these controllers existed (generated) but were never wired into the plugin; wiring them in is what makes the "served by currency_pair's own Qt surface" acceptance criterion literally true, not just aspirational.
4. Deleted all fx_convention files: Qt controller/dialogs/windows/ui, SQL create/drop scripts, and the ores.refdata api/core domain, repository, service, messaging, and generator files. Removed its wiring from the hand-maintained NATS registrar.cpp.
5. Discovered along the way that currency_pair's own SQL was never wired into the master =refdata_create.sql=/=refdata_drop.sql=/RLS policy scripts (generated in isolation by a prior task) — the tables didn't exist in the DB at all. Wired them in (create/drop order: classifications and groups before pairs, pairs before conventions; reverse on drop) since the migration is meaningless without them.
6. Discovered currency.spot_days (added to the model in an earlier commit but never regenerated) is needed to derive currency_pair.spot_days at read time, per the design in Reconcile currency_pair design with fx_convention. Regenerated only currency's domain/repository/generator/SQL-create facets (skipped Qt/CLI to avoid clashing with the concurrent "Commission: currency" story). A full SQL regen pulled in unrelated template drift (case normalisation, and a dropped tenant-fallback validation branch) so the SQL column additions were hand-applied instead of using the generated file wholesale.
7. Fixed an unrelated pre-existing build break in CurrencyDetailDialog.cpp (call to a generator function that no longer exists) as a one-line drive-by, since it blocked verifying the Qt wiring changes compiled.
8. Retired the ores.refdata.fx_convention codegen model doc (no incoming links) and removed it from ores.refdata.module.org's entity table, adding the currency_pair family in its place.

• Not done: the ores.refdata.api/core/qt PlantUML class diagrams (*.puml) still mention fx_convention / FxConvention — these are hand-authored/synced diagram sources, not generated from the code touched here, and updating them is a separate follow-up.
• Done: ores.refdata.service's NATS messaging registrar for currency_pair, currency_pair_convention, and currency_pair_classification — the nats-sub-registrar facet was written (mustache templates + two DONE tasks in the introduce-per-entity-sub-registrars story) but never actually added to facet_catalogue.org, so codegen couldn't produce these files and the three entities had handlers with no service wiring (NATS callers got "no service is handling subject"). Added the missing facet entry, tangled, generated the three registrar pairs via ores.cpp.nats-sub-registrar, and wired them into registrar.cpp. Verified live: a currency_pairs.list NATS request now returns unauthorized (routed and handled) instead of "no service is handling subject".
• Also moved the Qt Currency &Pairs menu action from Reference Data → Conventions (where it sat alongside actual curve-building conventions) to top level, next to Currencies — it is core reference data, not a convention.

fx_convention is fully retired: ores.ore, the Qt refdata plugin, and the CLI now read/write currency_pair=/=currency_pair_convention directly, with the divisor/multiplier pip_factor reciprocal computed inline rather than stored twice. Along the way, closed two gaps left by earlier tasks in this story — currency_pair's SQL was never wired into the master schema scripts, and currency.spot_days was modelled but never generated — both required for this migration to actually work end-to-end. Verified via a live DB recreate (tables present, fx tables gone) and a byte-for-byte ore roundtrip CLI run against a real ORE conventions.xml.