Task: Close the marketdata test coverage

Every marketdata source file has a test that exercises it, or a recorded reason, and the review items no one checked are recorded.

• V08: a per-file survey of src/ against the test run is on the task.
• Repository tests exist for feed_binding, observation_lineage and series_classification_rule; the generated handlers, publish_from_dq_handler, curve_republish_service and market_data_client are tested or recorded.
• Every entity has an eventing integration test, or a recorded reason.
• G04: the generated marketdata code has a review record.
• G07 and H04 are recorded.

Run the survey after the grammar cut-over, so the deleted files drop out of it.

Decided with the owner on 2026-10-05: the V08 survey is static. A script follows each test's includes through marketdata's headers and sources, and reports every src/ file with the tests that reach it. It measures reach, not line coverage; the instrumented build that would measure lines does not exist.

1. Turn on the generated eventing integration test for every entity that can take one.
2. Write repository tests for feed_binding, observation_lineage, series_classification_rule and the junction's list by series.
3. Move the as-of row mapping into as_of_rows and test its refusals.
4. Move on_tick's decision into plan_tick and test it.
5. Test the service's config parser.
6. Record the G04 review and the V08 survey, with a reason for every file no test reaches.

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.

Correct the marketdata clean-standard record recorded G07 and H04 as Pass at ec37b05f77. V08 and G04 are still open and stay with this task.

A name match finds no test for 73 of 102 source files. See the 2026-10-02 review, section 3, "Test coverage".

The review of PR 2607 found no test for feed_ingest_loop::on_tick. Cover these behaviours: a tick from an unbound source is dropped, a tick whose URI names no datum is dropped, a tick whose value is not a number is dropped, and a tick from a source with several bindings is stored and republished once for each binding.

The review of PR 2630 found no test for the as-of reads' refusals: a short row, a missing bucket ordinal, an ordinal that is not a number, and an ordinal outside the requested buckets. The row mapping sits inside the query call, so a test needs it split out first.

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.

Every marketdata entity with audit columns has an eventing integration test, and the two time-series entities are recorded with the template defect that stops theirs. The three repositories and the junction that had no repository test have one. The as-of reads' refusals and on_tick's decision moved into small functions that unit tests reach. The V08 survey is a script on the task: 65 of 114 source files are reached by a test, and each of the other 49 is recorded with its reason. The G04 review found six template defects, one a task and five a capture.