feat: Add ArrowBitsUnpackInt32()

[paleolimbot]

As a follow-up to #276. The int32 version is useful because R uses 32-bit integers to represent boolean (i.e., logical) arrays. This results in a significant speedup in boolean conversion!

@WillAyd: I updated a few things that you just added (Sorry! 😬 ):

• I changed Bitmap -> Bits and removed Unsafe to make it more consistent with the other functions that accept const uint8_t* bits
• I updated the test function so that it tests both the int32 and int8 types at once

Before this PR:

library(nanoarrow)

lgls <- nanoarrow:::vec_gen(logical(), 1e6)
bool_array <- as_nanoarrow_array(lgls)
bool_array_arrow <- arrow::as_arrow_array(bool_array)

bench::mark(
  convert_array(bool_array, logical()),
  as.vector(bool_array_arrow),
  as.logical(lgls)
)
#> # A tibble: 3 × 6
#>   expression                             min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                           <bch> <bch:>     <dbl> <bch:byt>    <dbl>
#> 1 convert_array(bool_array, logical()) 556µs  749µs    1.33e3    3.82MB     156.
#> 2 as.vector(bool_array_arrow)          558µs  780µs    1.30e3    3.82MB     144.
#> 3 as.logical(lgls)                         0    1ns    2.28e8        0B       0

bench::mark(
  convert_array(bool_array, integer()),
  as.integer(lgls)
)
#> # A tibble: 2 × 6
#>   expression                             min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                           <bch> <bch:>     <dbl> <bch:byt>    <dbl>
#> 1 convert_array(bool_array, integer()) 733µs  912µs     1093.    3.81MB     167.
#> 2 as.integer(lgls)                     615µs  788µs     1273.    3.81MB     182.

After this PR:

library(nanoarrow)

lgls <- nanoarrow:::vec_gen(logical(), 1e6)
bool_array <- as_nanoarrow_array(lgls)
bool_array_arrow <- arrow::as_arrow_array(bool_array)

bench::mark(
  convert_array(bool_array, logical()),
  as.vector(bool_array_arrow),
  as.logical(lgls)
)
#> # A tibble: 3 × 6
#>   expression                             min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                           <bch> <bch:>     <dbl> <bch:byt>    <dbl>
#> 1 convert_array(bool_array, logical()) 105µs  308µs    3.21e3    3.83MB     367.
#> 2 as.vector(bool_array_arrow)          559µs  772µs    1.30e3    3.82MB     143.
#> 3 as.logical(lgls)                         0      0    5.87e8        0B       0

bench::mark(
  convert_array(bool_array, integer()),
  as.integer(lgls)
)
#> # A tibble: 2 × 6
#>   expression                             min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                           <bch> <bch:>     <dbl> <bch:byt>    <dbl>
#> 1 convert_array(bool_array, integer()) 104µs  310µs     3181.    3.81MB     423.
#> 2 as.integer(lgls)                     615µs  784µs     1278.    3.81MB     142.

^{Created on 2023-08-17 with reprex v2.0.2}

[codecov-commenter]

Codecov Report

Merging #278 (e685462) into main (e21cc98) will increase coverage by 1.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##             main     #278      +/-   ##
==========================================
+ Coverage   87.22%   88.23%   +1.01%     
==========================================
  Files          66        3      -63     
  Lines       10158      357    -9801     
==========================================
- Hits         8860      315    -8545     
+ Misses       1298       42    -1256     

see 63 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[WillAyd]

no worries at all. I think these are all pretty reasonable - nice work!