[Enhancement]: zi_get_geometry() should validate county is a 5-digit GEOID, not silently return zero rows

[chris-prener]

Parent epic: #109

User story

As a repo maintainer/data consumer of zippeR::zi_get_geometry(), I want a clear, actionable error when county isn't supplied as a valid 5-digit GEOID, so that a malformed request (e.g. passing a county name string) fails loudly instead of silently returning a zero-row result.

Tier: tier/quick-win — low-effort (single input-validation check, matching the existing state/method validation style already in zi_get_geometry()), moderate-impact (prevents a confusing silent-failure UX gap)

Describe the bug

zi_get_geometry()'s county parameter is documented as requiring "a character scalar or vector with character GEOIDs (e.x. "29510")" (see R/zi_get_geometry.R roxygen @param county). If a user instead passes a human-readable county name (e.g. "St. Louis City"), the function does not validate the format and silently proceeds:

zi_get_geometry(year = 2020, state = "MO", county = "St. Louis City", method = "centroid")

Internally, zi_process_county() does counties <- counties[counties$GEOID %in% county, ], which matches zero rows when county isn't a real GEOID, and the function ultimately returns a zero-row sf object with no warning or error — easy to mistake for a package bug (as happened during UAT of #133/PR #138) rather than a usage error.

Expected behavior

zi_get_geometry() should validate that county values are 5-digit numeric-like GEOID strings (mirroring the existing state/method/year argument-checking style already in the function) and cli::cli_abort() with a clear message when they aren't, e.g.:

county must be a 5-digit character GEOID (e.g. "29510"). You provided "St. Louis City".

To Reproduce

zi_get_geometry(year = 2020, state = "MO", county = "St. Louis City", method = "centroid")
#> Simple feature collection with 0 features and 1 field   -- no error, silent zero rows

# correct usage (should be the only accepted form):
zi_get_geometry(year = 2020, state = "MO", county = "29510", method = "centroid")
#> Simple feature collection with 19 features and 1 field  -- works as expected

Desktop (please complete the following information):

• OS: macOS
• Version of R: 4.4.1
• Version of RStudio: N/A

Additional context

• Confirmed empirically: with the correct GEOID "29510", both method = "centroid" and method = "intersect" return the expected non-zero row count. The zero-row result was a UAT usage error, not a regression from refactor: convert dplyr rename/rename_with/bind_rows and remaining pipeline-glue call sites to Base R #133 — filing this purely as an ergonomics/validation gap.

Codebase Context

• R/zi_get_geometry.R: zi_get_geometry() (~line 122) already validates year, style, class, return, shift_geo, and method with explicit cli::cli_abort() checks near the top of the function — county has no equivalent check and should follow the same pattern.
• R/zi_get_geometry.R: zi_process_county() (~line 435) is where the silent zero-row result actually originates (counties[counties$GEOID %in% county, ]).
• Related: [Epic H] User-Facing API Ergonomics & Dependency Reduction #109 (Epic H, User-Facing API Ergonomics & Dependency Reduction) — this is exactly the kind of input-validation ergonomics gap that epic tracks.