Story: Clean ores.scheduler to the component clean standard

ores.scheduler meets the Component Clean Standard: every item: All, Protocol and Entity pass, each item that does not apply is recorded with its reason, and ores.scheduler joins COMPONENTS_UNDER_TEST. Order 20 of 35 in the sprint 26 component clean-up, tier 2 (higher infrastructure).

• Every checklist item that applies to a component of kind Entity passes, with its evidence recorded on the task.
• Every item that does not apply is recorded as not applicable, with the reason.
• Every generator defect found is fixed at its source, with a regression test.
• ores.scheduler passes its own gate, check_component_drift.py --component scheduler-cpp. It joins COMPONENTS_UNDER_TEST, and with it the shared --all list, in the job_instance unit (V06); the shared list deliberately covers only a component whose regeneration leaves the tree fully clean.

• Survey on 2026-09-23: 1 entity, junction or operation models; 62 C++ files outside tests, 0 of them generated (0%); 5 messaging headers, 5 of them hand-written.
• Nothing is generated yet. Known work sits in the inbox story "Complete the LLM skills catalogue": the scheduler audit findings, the hand-written operations to model, and the custom history handler to replace. Take those tasks into this story.
• Survey on 2026-09-26, before any edit: 39 public headers, reproduced by command at f7fc9396b7; nine of the names they declare had a production consumer outside the component, all of them in ores.reporting, and that list is verified by git grep at the same revision: cron_expression, its rfl::Reflector, job_definition, and the three request and response pairs of the old protocol. The survey also tallied 59 declared symbols, a manual count that no command reproduces – counting named types alone gives 45 – so the task keeps the verified list rather than a ratio. One model was visible to codegen, eight subject literals were in code, and regeneration would have rewritten 19 files and created 12.
• The job_definition unit is done: the model is resynced and bound to uuid-identified-lookup, the entity stack is generated, the retired schedule, unschedule and schedule-batch subjects are gone in favour of the canonical put, put_many and delete verbs, ores.reporting has moved to them, and the generated history provider replaces the custom history handler. Five generator defects were found and fixed at source, each with a regression test.
• The units that remain are recorded on the task, each with its size: the job_instance model and the registry entry (M06, V06 and V05's last clause), the scheduler shell part and its commands (S01, S02 and V04), the scheduler loop's I/O path – the decisions it computes are tested now, the database and NATS work is not – the model-property reference gap, the repository-wide component_files.cmake sweep, the drift check's repository-wide diff, the SQL ER diagram that no gate covers, and the diagram generator's inability to regenerate a composite component's modules.