Task: Move the swap leg's notionals and rates into child rows

Give a swap leg's notionals and rates child rows keyed by the leg, populate them from the full ORE lists, and stop the flat columns from standing in for a schedule.

• ores_trading_swap_leg_amounts_tbl and ores_trading_swap_leg_rates_tbl exist, keyed by the trade, the leg ordinal and an entry ordinal.
• swap_leg no longer carries notional, fixed_rate or spread; the children carry them.
• map_leg reads every notional and every rate or spread, not the first of each.
• The import writes the children and the export reads them back, so a stepped leg round trips.
• The component regenerates drift-clean and builds, and the codegen suite passes.

The leg's economics become rows of their own and the wire carries them, in five parts.

1. Two child models, swap_leg_amount and swap_leg_rate, keyed by the trade, the leg ordinal and their own ordinal, with rate_role naming the arm that stated the rate (fixed or spread).
2. swap_leg loses notional, fixed_rate and spread: a single column cannot hold a schedule, and keeping both would state one fact twice.
3. map_leg stops taking the first of each list; map_leg_amounts and map_leg_rates read every entry, including the Amortizations arm when the document states a schedule instead of a list.
4. The carrier swap_instrument_data gains the two collections, and the hand-written payload codec carries them, so the export can rebuild the lists it reads.
5. The import writes the children after the legs, and the export reads them back through a batch read on each child.

The children are flat, like the instrument_* children beside them: no identity field group and no audit group, so their domain class is one struct.

Every number this task claims about a tree, with the command that produced it and the commit it was measured at.

The children's tables are new, so their permission codes had to be seeded in iam_permissions_populate.sql; the handler-permission gate fails without them. The component's component_files.cmake lists also had to be regenerated, or the new repository units are never compiled and the link fails.

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

A leg's notionals and its rates are rows of their own, and the import keeps every one of them.

• Two child tables. swap_leg_amount holds one notional per row, dated from the day it applies; swap_leg_rate holds one rate or spread, with rate_role naming the arm of the document that stated it. Both are keyed by the trade, the leg ordinal and their own ordinal.
• swap_leg lost its economics columns. notional, fixed_rate and spread are gone, because a single column cannot hold a schedule and keeping both would state the same fact twice.
• The mapper keeps the whole list. map_leg read first_notional(...), Rates.Rate.front() and Spreads.Spread.front() and dropped the rest; the new map_leg_amounts and map_leg_rates read every entry, and read the Amortizations arm when the document states a schedule rather than a list. The amortisation's own type, end date and frequency are not carried yet, and the reason is that this row states a value and a start date.
• The children cross the wire. swap_instrument_data gained the two collections and the hand-written payload codec carries them, with a fallback tier for a payload written before the change, because the export rebuilds a leg from the payload and nothing else.
• The import writes them and the export reads them back, through a batch read on each child that mirrors the legs' own.

Verification: the component regenerates with no drift; the codegen suite is 1276 passed; the full preset build exits 0 with no errors and no compiler warnings; and the ORE mapper round-trip suite is 614 cases and 482159 assertions, all passing, against a database recreated from this branch. That last one is the proof that matters: the golden IR and swaption documents state stepped legs, the export rebuilds them from the children, and the comparison is byte-level.

Not run: the trading service's eventing integration tests. They need the service fleet, and this environment's user systemd bus is down, so services cannot start. The database was recreated for the mapper suite, which needs no fleet.