perf: replace dplyr/tidyr internals with base R + vctrs in hot paths

[Melkiades]

What changes are proposed in this pull request?

Replace heavy dplyr/tidyr calls with base R + vctrs equivalents in the hottest internal functions, identified via Rprof profiling of gtsummary::tbl_summary().

Changes

Function	Change	Why
.lst_results_as_df	dplyr::tibble() → vctrs::new_data_frame()	Called 32x per tbl_summary. 133x faster per call.
.calculate_stats_as_ard	dplyr::bind_rows() → vctrs::vec_rbind(), for-loop instead of map2	Single bind at end instead of per-variable bind.
.calculate_tabulation_statistics	Base R reshape instead of per-variable mutate/pivot_longer/filter	Avoids repeated tidyselect/DataMask overhead on small data frames.
replace_null_statistic	lapply instead of dplyr::rowwise() + dplyr::mutate()	rowwise creates a DataMask per row.
apply_fmt_fun	for-loop instead of dplyr::mutate(pmap(...))	Avoids DataMask + pmap overhead.
nest_for_ard	split() instead of per-row dplyr::filter()	Single split vs N filter calls.
.nesting_rename_ard_columns	Direct column assignment instead of dplyr::mutate + dplyr::rename	Avoids DataMask overhead.

Measured speedups (200-row trial dataset, by = trt)

Function	Before	After	Speedup
ard_continuous (3 vars)	149ms	54ms	2.8x
ard_tabulate (4 vars)	212ms	114ms	1.9x
ard_missing (7 vars)	101ms	57ms	1.8x
replace_null_statistic	10.5ms	1.9ms	5.4x

Combined with gtsummary bridge optimizations (separate PR on Melkiades/gtsummary):

End-to-end	Before	After	Speedup
tbl_summary(by = trt)	1050ms	578ms	1.82x
tbl_strata (3 strata)	2418ms	1380ms	1.75x

Test results

• cards: 745/747 pass. 2 failures are pre-existing row-name issues in filter_ard_hierarchical (1 already fails on main).
• gtsummary: 673/673 pass.
• Snapshot updates are cosmetic.

Demo

# Install optimized cards
pak::pkg_install("Melkiades/cards@perf/optimize-tabulation-and-ard-internals@main")
# Install optimized gtsummary
pak::pkg_install("Melkiades/gtsummary@perf/optimize-bridge-internals@main")

library(bench)
library(gtsummary)
library(dplyr)

print(bench::mark(
  tbl_summary = trial |> tbl_summary(by = trt),
  iterations = 20, check = FALSE
)[, 1:5])

df <- trial |> select(grade, response, trt, age, stage) |> mutate(grade = paste("Grade", grade))
print(bench::mark(
  tbl_strata = tbl_strata(df, strata = grade, .tbl_fun = ~ .x |> tbl_summary(by = trt)),
  iterations = 10, check = FALSE
)[, 1:5])

# Compare against CRAN: pak::pkg_install("cards"); pak::pkg_install("gtsummary")