Attach StatCan geographies to geocode() / reverse_geocode() results

Split out of #1, which is closed: `Locations` is imported and carries `CSD_CODE`, `FED_CODE`/`FED_ENG_NAME`/`FED_FRE_NAME` and `ER_CODE`/`ER_ENG_NAME`/`ER_FRE_NAME`, joined to `Addresses` on `LOC_GUID`.

What is missing is a convenience layer. Today a user who wants the census subdivision or federal riding for a geocoded address has to know the schema and write the join themselves — the Querying NAR vignette shows how, but it is manual:

tbl(con, "Addresses") |>
  left_join(tbl(con, "Locations") |> select(LOC_GUID, FED_ENG_NAME, ER_ENG_NAME),
            by = "LOC_GUID")

Proposal: attach the geography for callers who never touch the database directly.

Open questions, roughly in the order they need answering:

• Shape. An argument on the existing functions (`geocode(geographies = TRUE)`, `reverse_geocode(geographies = ...)`) or a standalone verb (`nar_geographies(x, con)`) that takes a result and adds columns? A standalone verb composes better and keeps the two entry points from growing a third mode each.
• Which columns, and named how. All of CSD/FED/ER, or a selectable subset? NAR ships codes and both official-language names; returning nine columns by default is a lot.
• Rows with no `ADDR_GUID`. Interpolated and `bc_*` rows have no NAR record to join through, so the geography would have to come from the point rather than the key — a spatial lookup against something the package does not currently carry. Simplest first cut is to fill `NA` for those and document it; anything better needs boundary files.
• `geom_source`. A blockface-derived point can sit on the far side of a boundary from the building it stands for. Worth a note if the geography is ever derived spatially.

Start with geocode(), reverse_geocode(), the Querying NAR vignette, and the Addresses-to-Locations join on LOC_GUID. Resolve the proposed API shape, geography columns, handling of rows without ADDR_GUID, and geom_source semantics; done means the chosen interface attaches documented geography results without requiring callers to write the join.