Keep filter catalogs independent from filtered analytics datasets

Observed 2026-09-04 while reviewing theft PR #723 / COR-3895. In a faceted analytics UI, do not build an independent dimension catalog (for example, broker choices) from the dataset narrowed by another active facet (for example, Created Via). Doing so can erase the currently selected dimension and make client normalization broaden results back to All. Preserve an unfiltered, authorization-scoped catalog for option validation, while separately querying the filtered rows used for KPI calculations. For large joins where attribution exists only on flows, first apply authorization + attribution in SQL, then perform bounded indexed lookups for related reference rows; avoid fetching all organization flows for in-memory filtering. Provenance: native review findings and local code inspection of web/src/lib/shipper-analytics-page-data.ts and web/src/lib/shipper-reference/references.ts at local HEAD 8a8cd401; publication remained blocked pending the catalog/data split.