Drop the inert gdp-population area relabelling

[lbm364dl]

What was wrong

.canonicalise_gdp_pop_area() (R/build_cbs.R) relabelled the
gdp-population pin's area column into the polity-name vocabulary. Its
30-line comment and its two tests both said why:

.fill_with_proxies() joins population on c("year", "area") -- the name, not
a code -- and the two sides speak different vocabularies.

That stopped being true at commit 2210d05d. .fill_with_proxies() now does

pop_dt <- .proxy_bucket_key(gdp_pop, iso3_col = "area_code")
pop_dt <- pop_dt[, .(pop = sum(pop, na.rm = TRUE)), by = .(year, area_code)]
dt <- merge(dt, pop_dt, by = c("year", "area_code"), all.x = TRUE, sort = FALSE)

.proxy_bucket_key() keys on the ISO3 in area_code and the aggregation
reduces the pin to (year, area_code, pop), so the area label is dropped
before it reaches any join. Verified .fill_with_proxies() is the only
consumer of .read_gdp_pop()'s output (gdp_pop flows
.read_cbs_inputs() → inputs$gdp_pop → .fill_historical_destinies() →
.fill_with_proxies(), and nothing else reads it); get_population() in
R/population.R reads the pin directly via whep_read_file(), not through
.read_gdp_pop(). So the relabelling was inert with respect to every
published number, and the tests would have kept passing had the function been
identity().

What changed

Option 1 from the issue. Deleted the function, its stale 30-line comment
block, its two tests, and the two now-unused utils::globalVariables()
entries (canonical_area, key_row — grepped, used nowhere else).
.read_gdp_pop() now only renames Year → year, with a short comment
recording why the label needs no canonicalisation and which commit made it
so.

Two tests replace the deleted pair:

• .read_gdp_pop renames only the year column — mocks .read_input() with a
  four-row fixture and asserts the pin's area comes back untouched. This is
  the behaviour change, and it fails on the old code.
• .fill_with_proxies ignores the population pin's area label — the guard
  that makes the removal safe: garbling the pin's area to
  "not a polity name" leaves both the filled food series and the joined
  pop identical. If a proxy is ever keyed on the name again, this fails
  loudly instead of the fill silently going unfilled.

No published value changes — the removed code only ever wrote a column
that was aggregated away. Classification: mechanical (dead-code removal
plus a false comment; no methodological choice).

Verification

Failing before the fix (source stashed, new tests in place):

── Failed tests ────────────────────────────────────────────────────────────────
Failure ('test_build_cbs.R:1216:3'): .read_gdp_pop renames only the year column
result$area (`actual`) not identical to pin$area (`expected`).

`actual`:   "Laos" "South Korea"       "Albania (1913-2025)" "Spain"
`expected`: "Lao"  "Republic of Korea" "Albania"             "Spain"

[ FAIL 1 | WARN 6 | SKIP 0 | PASS 148 ]

Passing after:

══ Results ═════════════════════════════════════════════════════════════════════
Duration: 3.3 s

[ FAIL 0 | WARN 6 | SKIP 0 | PASS 149 ]

(The 6 warnings are the pre-existing unfold_rest_of_world notices, unchanged
by this PR.) air format . made no further changes; lintr clean on all
three touched files (R/build_cbs.R, R/utils.R,
tests/testthat/test_build_cbs.R). No roxygen changed, so man/ and
_pkgdown.yml are untouched. Everything here is offline — no pin is read.

Closes #721.

🤖 Generated with Claude Code