fix(r): Fix map type and struct-nested-in-struct convert

[paleolimbot]

Some of the types listed in the Overture maps foundation Parquet files are deeply nested in various ways we hadn't encountered yet when trying to convert to R.

The motivating reprex here was:

library(sedonadb)

sd_read_parquet("/Volumes/data/overture/data/theme=divisions/type=division_area/") |>
  head(100000) |> 
  sd_collect() |> 
  tibble::as_tibble()
#> Can't convert `common` <map<key_value: struct<key: string, value: string>>> to R vector of type vctrs_list_of
#> or
#> Something about expected length mixing up rows and columns of a data frame

After this PR:

library(sedonadb)
#> Warning: package 'sedonadb' was built under R version 4.5.2

sd_read_parquet("/Volumes/data/overture/data/theme=divisions/type=division_area/") |>
  head(100000) |> 
  sd_collect() |> 
  tibble::as_tibble()
#> # A tibble: 100,000 × 13
#>    id     geometry bbox$xmin country version sources subtype class names$primary
#>    <chr>  <grrw_v>     <dbl> <chr>     <int> <list<> <chr>   <chr> <chr>        
#>  1 194bd… <POLYGO…     -16.6 ES            1 [1 × 7] microh… land  "Plaza Paco …
#>  2 f7334… <MULTIP…     -16.6 ES            3 [1 × 7] locali… land  "Los Realejo…
#>  3 a273d… <POLYGO…     -16.6 ES            1 [1 × 7] microh… land  "Plaza de la…
#>  4 46d32… <POLYGO…     -16.6 ES            1 [1 × 7] microh… land  "Plaza de la…
#>  5 897c5… <POLYGO…     -16.6 ES            1 [1 × 7] microh… land  "Plaza Domín…
#>  6 92c7c… <POLYGO…     -16.6 ES            1 [1 × 7] microh… land  "Plazoleta M…
#>  7 afdbf… <POLYGO…     -16.6 ES            1 [1 × 7] microh… land  "Plaza Poeta…
#>  8 8ed3a… <POLYGO…     -16.6 ES            1 [1 × 7] microh… land  "Plaza de Pe…
#>  9 319f4… <POLYGO…     -16.7 ES            2 [1 × 7] locali… land  "La Guancha" 
#> 10 36457… <MULTIP…     -16.7 ES            3 [1 × 7] locali… land  "San Juan de…
#> # ℹ 99,990 more rows
#> # ℹ 9 more variables: bbox$xmax <dbl>, $ymin <dbl>, $ymax <dbl>,
#> #   names$common <list<df[,2]>>, $rules <list<df[,6]>>, is_land <lgl>,
#> #   is_territorial <lgl>, region <chr>, division_id <chr>

Interestingly, nanoarrow seems to be much faster than arrow at converting this deeply nested structure. I have no idea why.

complex_map_file <- system.file("test-data/complex-map.arrows", package = "nanoarrow")

bench::mark(
  arrow = as.data.frame(arrow::read_ipc_stream(complex_map_file)),
  nanoarrow = as.data.frame(nanoarrow::read_nanoarrow(complex_map_file)),
  check = FALSE
)
#> # A tibble: 2 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 arrow       44.95ms  44.95ms      22.2   21.04MB    178. 
#> 2 nanoarrow    3.97ms   5.81ms     178.     2.06MB     23.8

^{Created on 2025-12-05 with reprex v2.1.1}