Task: Repair the swap leg children's insert trigger and isolation

ores_trading_swap_leg_amounts_tbl and ores_trading_swap_leg_rates_tbl accept a row again, and are isolated to their tenant. The two tables bind a profile that describes a single-row instrument, so their generated insert trigger stamps a party_id column their flattened shape does not have, and every insert fails. Their siblings carry the tenant isolation policy they are missing.

• Neither generated create file stamps a column its table does not declare.
• Both tables enable row level security and carry a tenant isolation policy, as ores_trading_bond_leg_amounts_tbl does.
• Both drop files drop that policy, and the drop aggregator reaches all four drop files.
• The tables' column shape is unchanged, so no data migration is implied.

1. The root cause is the :profile: trading-instrument binding. The loader resolves the profile into two features, has_tenant_id and party_id_from_session, and the second emits the party stamp. The profile describes a flat ORE instrument sub-type with an identity group that carries party_id; a leg's notional row is not an instrument sub-type and has no such column.
2. Replace the binding with the one feature the tables actually need, :has_tenant_id: true, which is exactly what the working sibling ores.trading.bond_leg_amount declares. The stamp disappears and nothing else about the trigger changes.
3. Add :rls_tenant_isolation: true to both models' SQL Flags, which generates the enable and the tenant policy into each create file and the policy drop into each drop file.
4. Wire the four drop files into drop_trading.sql, children before the leg they belong to, because the drop aggregator never reached them.
5. Regenerate the component and confirm the trigger change is the only difference to the tables.

The warnings that remain belong to other families: ores_trading_structures_tbl (one RLS_001 and one RLS_002), ores_marketdata_market_series_identity_tbl (one RLS_002 and one TENANT_002), and the unwired drop files of the structures and trade link families.

• The defect was latent: no test writes a swap leg amount or rate row yet, so the round trip passed while the table could not accept one.
• swap_legs, the parent, has a party_id column and a hand-written policy in trading_rls_policies_create.sql, which is why it was unaffected. Its children have neither.
• The regeneration also refreshed structure_service.cpp, whose committed copy the outcome-catalogue template had outrun. It is generated output, so the gate requires the regenerated form.

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.

Both tables accept a row and are tenant-isolated. The profile binding that described a single-row instrument is gone from two child tables that are not instruments, the tenant isolation policy matches the bond family's children, and the drop aggregator reaches the four drop files. The table shape is unchanged.