Task: Migrate currency.holiday_calendar to a real FK relationship

Replace currency.holiday_calendar's free-text value with a proper FK relationship to the new calendar entity, via a junction table so a currency can reference more than one calendar (e.g. a country calendar and a financial-centre calendar).

• [X] A currency_calendar (or similarly named) junction table exists, FK-referencing both currency and calendar.
• [X] currency.holiday_calendar's existing free-text values are migrated onto the junction table.
• [X] The old free-text column is retired.

Mirrored the exact approach used for the currency_country junction: codegen for SQL only (ores.codegen.junction), hand-authored C++ layer copied from currency_country and re-targeted from country_alpha2_code to calendar_code. Retired currency.holiday_calendar via a normal codegen regeneration of the currency entity (field deleted from the org model, then compass codegen entity generate currency), which cleanly dropped the column everywhere except the two places codegen doesn't own: the DQ publish-from-dq function (stopped writing the retired column) and presentation_currency_history_field_mapper_tests.cpp's hardcoded field count.

Modelled currency_calendar as an ores.codegen.junction (SQL schema generated) and hand-authored the C++ layer mirroring currency_country exactly (domain struct, JSON I/O, table display/table I/O, custom generator, repository entity/mapper/repository), since junction codegen is SQL-only. Wired the create/drop scripts into refdata_create.sql=/=refdata_drop.sql (currency_calendar depends on both currencies and calendars, so it creates after both and drops before both) and added its RLS tenant-isolation policy up front, having just fixed that same class of gap for calendars=/=calendar_types and currency_countries on main.

Seeded 19 real currency-to-calendar mappings (mirroring the values previously hardcoded on currency.holiday_calendar, e.g. EUR -> TARGET, USD -> UnitedStates.Settlement, GBP -> UnitedKingdom.Settlement) via both a generate_currency_calendars() custom generator and a matching SQL populate script (refdata_currency_calendars_populate.sql), following the same dual-path seed pattern established for calendar's QuantLib seed. Several currencies needed a QuantLib sub-market variant rather than the bare country token (UnitedStates.Settlement, Canada.Settlement, UnitedKingdom.Settlement, China.SSE, SouthKorea.Settlement), since the seeded calendar set has no plain root token for those markets.

Retired currency.holiday_calendar by removing the field from ores.refdata.currency.org and regenerating the currency entity via codegen, which cleanly dropped the column from the domain struct, entity, mapper, generator, table/table_io, Qt model, and SQL create script. Fixed the two non-codegen-owned consumers: stopped the publish-from-dq function writing the now-retired column (still reads the raw value from the DQ artefact table, unchanged, since that's upstream staging data outside this task's scope), and updated presentation_currency_history_field_mapper_tests.cpp's hardcoded field count (21 -> 20). refdata_currency_pair_conventions_seed_populate.sql reads holiday_calendar off the DQ artefact table directly (not the retired refdata column), so needed no change.

Verified end-to-end: full monorepo build clean; ctest 72/73 passed (the 1 failure is the same pre-existing, unrelated contact_type generator test seen throughout this sprint, confirmed untouched by any commit here); validate_schemas.sh clean (0 warnings, RLS_001 included); compass db recreate runs clean with 19 currency_calendars rows seeded.