Task: Retire the unproduced rpa_instruments table

The unproduced rpa_instruments table and everything generated from it are gone, and the tree builds and passes without them.

• The ores.trading.rpa_instrument model, its generated C++, SQL, shell commands and TypeScript are removed, 47 tracked files at the time of the analysis.
• The rates wire variant no longer names an rpa_instrument leaf.
• The inbox capture for the retirement is closed.
• The database recreates from the create scripts, codegen drift is clean, and the build and tests pass.

1. Census. List every file named for the table and every file that mentions it, under every spelling: snake case, camel case and kebab case.
2. Remove the named files. The model and everything generated from it: the domain type, the repository, the service, the messaging, the eventing, the handler, the generator, the history mapper, the SQL create and drop scripts, the shell commands, their scripts and tests, the TypeScript protocol and the recipe. Keep the task and the capture pages.
3. Remove the references. The RLS policies, the permission seeds, the create and drop includes, the module listing, the event and handler registrations, the shell aggregator, the recipe index and the five component source lists. Regenerate what a script owns.
4. The diagrams. Strip the table from the five diagram sources and re-render the committed images.
5. The capture. File the inbox capture as discarded with a note that this task did the work, and regenerate the backlog indexes.
6. Prove it. Build the whole tree, recreate the database from the create scripts, and run the suites of every component the unit touched.

Every number this task claims about a tree, with the command that produced it and the commit it was measured at. A number without both is not a measurement, because a clean-up record describes a tree and trunk moves under a branch: a consumer count, a gate's component count and a suite's size all change with no code changing, and the reader of a bare number cannot tell which tree it came from. Where the raw output matters, put it in an evidence/ file beside the task and name that file in the Command column.

• Three files outside doc/agile still mention the table, and all three are historical: two archived plans under doc/plans and a prototype page that holds a snapshot of the permission list. They are left as they are.
• The capture named an rpa_instrument leaf of the rates wire variant. After the removal a search of projects finds no file that mentions the table, so no variant names it.
• The two ORE suites ores.ore.service.tests and ores.ore.api.tests hold no test cases, so there was nothing to run in them.

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.

The unproduced rpa_instruments table and everything generated from it are gone. 47 files named for it are deleted, and the references in the SQL create, drop, policy and permission scripts, the registrations, the shell aggregator, the recipe index, the module listing, the catalogue note, the five component source lists and the five diagrams are removed. The inbox capture is filed with a note that this task did the work.

The whole tree builds, the database recreates from the create scripts, and the trading, ORE, shell, trading API and trading service suites pass. The risk participation agreement still routes to the credit table, as before; its own facts get a table in unit 4.

Acceptance met: the model and its generated files are removed, the rates wire variant no longer names the leaf, the inbox capture is closed, and the database recreates, the build and the tests pass. The drift check is run on the committed tree before the PR.