Some speedup for *_inner_join()

[etiennebacher]

Hi @beniaminogreen, here's a small PR that slightly improves the performance of *_inner_join() functions. For those functions, we don't need to get the indices of rows that don't match since they are not used in the output anyway. Small benchmark below (about 15% time decrease):

library(bench) 
library(dplyr)

out <- cross::run(
  pkgs = c("beniaminogreen/zoomerjoin", "etiennebacher/zoomerjoin@speedup-inner-join"), 
  ~{
    library(zoomerjoin)
    
    ### Setup -----
    
    corpus_1 <- data.table::rbindlist(rep(list(dime_data), 300))
    names(corpus_1) <- c("a", "field")
    print(nrow(corpus_1))
    
    corpus_2 <- data.table::rbindlist(rep(list(dime_data), 300))
    names(corpus_2) <- c("b", "field")
    
    ### Benchmark -----
    
    bench::mark(
      jaccard_inner_join(corpus_1, corpus_2, 
                        by = "field", n_gram_width=6,
                        n_bands=20, band_width=6, threshold = .8)
    )
  }
)

tidyr::unnest(out, result) |> 
  select(pkg, median, mem_alloc)
#> # A tibble: 2 × 3
#>   pkg                                           median mem_alloc
#>   <chr>                                       <bch:tm> <bch:byt>
#> 1 beniaminogreen/zoomerjoin                       3.7m    15.7GB
#> 2 etiennebacher/zoomerjoin@speedup-inner-join    3.21m    10.3GB

(the cross allows to run the same code on different branches, which is very useful for benchmarking)

---

Also worth noting that when getting the indices of unmatched rows is necessary, using collapse::%!iin%`` instead of !%in% would be more efficient (but would also add a dependency to the package)

[beniaminogreen]

Thanks for this! This is a great catch! Merging in now.

The cross function is a great tip as well - I have always struggled with testing across branches.