Drop dead .polity_bridge() and mock the crosswalk the aggregator uses

[lbm364dl]

What was wrong (mechanism)

.polity_bridge() (R/read_raw_inputs.R) memoised an area → polity bridge that
nothing called. Confirmed against the loaded namespace rather than by grep,
so a call assembled as a string would also have shown:

ns <- asNamespace("whep")
Filter(function(nm) {
  f <- get(nm, envir = ns)
  is.function(f) && any(grepl("\\.polity_bridge", deparse(body(f))))
}, ls(ns, all.names = TRUE))
#> character(0)

.aggregate_to_polities(), the only plausible caller, resolves through
.add_polity_columns_dt() → .polity_crosswalk() instead, because it needs the
year-aware lookup; .polity_bridge() wrapped the year-blind
.current_area_lookup() (one row per area_code).

The cost was in the tests: two blocks in test_read_raw_inputs.R mocked
.polity_bridge() and read as though the fixture controlled what the aggregator
resolved. It did not. They passed because the shipped crosswalk resolves area
203 to ESP-1800-2025 / "Spain" / bucket 203 on its own — so they were not the
hermetic unit tests they looked like, and a reader reasoned from a fixture that
was never in play.

What changed

• Deleted .polity_bridge().
• Replaced the two inert local_mocked_bindings(.polity_bridge = …) blocks with
  a .local_aggregator_crosswalk() helper that mocks .polity_crosswalk() —
  the binding the aggregator actually reaches — mirroring what
  test_polity_folds.R's .local_fold_crosswalk() already does. Its fixture is
  deliberately not a shipped value (Testland / TST-1900-2025).
• Added a test asserting the aggregator carries the fixture's label and reporting
  polity code out, so an inert mock cannot pass again.

Classification: mechanical — dead-code removal plus a test-fidelity fix. No
published value changes; the deleted helper had no callers.

Verification

The new assertion fails whenever the mock is inert. Probe run on the unfixed
code (git stash of the source change), mocking .polity_bridge() with the same
Testland fixture:

✖ | 2        0 | bridge_probe
Failure ('test_bridge_probe.R:22:3'): mocking .polity_bridge controls the aggregator (whep#590 probe)
result$area (`actual`) not equal to "Testland" (`expected`).
`actual`:   "Spain"
`expected`: "Testland"

Failure ('test_bridge_probe.R:23:3'): …
result$reporting_polity_code (`actual`) not equal to "TST-1900-2025" (`expected`).
`actual`:   "ESP-1800-2025"
`expected`: "TST-1900-2025"

[ FAIL 2 | WARN 0 | SKIP 0 | PASS 0 ]

The shipped Spain values leaking through the fixture are the defect. With the
fix, the same assertions routed through the mocked .polity_crosswalk() pass:

devtools::test(filter = "read_raw_inputs")
[ FAIL 0 | WARN 0 | SKIP 0 | PASS 26 ]

devtools::test(filter = "polity_folds")   # aggregator's other consumer
[ FAIL 0 | WARN 11 | SKIP 0 | PASS 114 ]

air format . clean; lintr clean on both changed files (no lints). No roxygen
change (the helper was undocumented), so no man/ or _pkgdown.yml update.

Closes #590.

🤖 Generated with Claude Code