Validate .rows in tibble()

[sims1253]

tibble_quos() in R/tibble.R copies .rows into first_size without checking it. The first evaluated column then hits first_size == 1L && current_size != 1L, so a vector .rows fails inside && with a base R message, and other bad values surface as errors about something else:

tibble(a = 1:3, .rows = c(1L, 5L))
#> Error: 'length = 2' in coercion to 'logical(1)'
tibble(a = 1:3, .rows = 1.5)
#> Error: Tibble columns must have compatible sizes.
tibble(.rows = -1)
#> Error: `nrow` must be a nonnegative whole number smaller than 2^31.

This PR validates .rows once, before the column loop. It must be NULL or a single non-negative whole number. Anything else goes through a new abort_bad_rows_arg() helper that follows the existing tibble_abort(problems(...)) pattern, so the condition gets the class tibble_error_bad_rows_arg.

All three calls now fail the same way, with the offending value on the second line:

Error in `tibble()`:
! `.rows` must be `NULL` or a single nonnegative whole number.
x `.rows` is c(1L, 5L).

Found during my most recent ry audit.