Parameterize api-core's cross-entity test duplication

[chet]

api-core re-proves the same behavior once per entity, which inflates the suite without adding coverage. Two patterns dominate: every find_*_by_ids RPC has its own "at least one id" / "no more than N ids" guard test (~18 across 10+ entities), all asserting byte-identical messages; and the health-override CRUD suite is implemented fn-for-fn three times (power_shelf, switch, rack). This consolidates both. The broader #[sqlx_test] integration tests that exercise distinct per-entity logic are out of scope -- they legitimately stay separate.

This is the heaviest item in Wave 1; if the diff gets unwieldy it can split into two PRs (guards; CRUD), with a linked issue opened for the second.

Scope

• Guards: replace the per-entity *_by_ids_none / *_by_ids_over_max tests (tests/*_find.rs) with one canonical test per guard, or a parameterized table keyed on a per-RPC closure. The guard runs in the API layer ahead of the DB, so it does not need re-proving per entity.
• Health-override CRUD: parameterize the power_shelf_health / switch_health / rack_health suites (tests/) into one generic harness over (insert/list/remove/find fns, metric prefix, alert id); keep rack_health's unique propagation/precedence tests.
• component_manager: collapse the power_action_* (6) and error_to_status_* (5) per-variant tests into tables (handlers/component_manager.rs) -- the latter also adds the currently-missing Transport / Rms arms.

Acceptance

• Each guard is proven once; the health CRUD suite runs through one harness; coverage preserved or improved (component_manager gains Transport / Rms).
• cargo test green for api-core; cargo clippy --all-features clean.

Size: M-L

Part of #2692.