plot.R: non-NA row-count filter collapses to a scalar, giving wrong point nudge spacing

The bug is in R/plot.R, lines 211, 500, and 852, within the functions plot_dot(), plot_errorbar(), and table_panel(). Start by reading the current row-count filter logic and the provided reproducer. The fix is to replace the scalar sum with rowSums to filter rows not entirely NA. Test the change by running the reproducer and checking the static RTF plotting path (rtf_static_forestly()).

In plot_dot(), plot_errorbar(), and table_panel() the per-item vertical nudge is computed from a "count of rows that are not fully NA" (R/plot.R, e.g. lines 211, 500, 852):

row_count <- max(nrow(x[!sum(is.na(x[, 1:n_col])) == n_col, ]), 1)

The index expression does not do what it looks like. sum(is.na(x[, 1:n_col])) sums NAs over the entire sub-frame, producing a single scalar; !scalar == n_col is therefore one logical value, which is recycled across all rows. So the filter keeps all rows or zero rows, never a per-row subset.

Reproducer (n_col = 2):

sel <- function(df) { keep <- !sum(is.na(df[,1:2])) == 2; nrow(df[keep,,drop=FALSE]) }
sel(data.frame(x1=c(1,2),  x2=c(3,4)))    # 2  (no NA -> keep all)
sel(data.frame(x1=c(1,NA), x2=c(3,NA)))   # 0  (2 NAs total -> drop all)

Effect: row_count equals the full group size in the normal case, but is forced to 1 whenever a group happens to contain exactly n_col NA values total, which changes nudge_unit(row_count) and misplaces the plotted points vertically. This only affects the static RTF plotting path (rtf_static_forestly()), and the visual effect is subtle, but the logic is incorrect.

Likely intended: count rows that are not entirely NA, per row, e.g.

not_all_na <- rowSums(is.na(x[, 1:n_col])) != n_col
row_count  <- max(sum(not_all_na), 1)

The same expression is duplicated in all three functions and should be fixed once (see also the broader duplication across these functions).