`round.ITime` fails when given numeric `digits` by `xfun::md_table`

[aitap]

Found while pre-rendering examples for the website:

data.table::as.ITime('23:58:00') |> data.frame(x=_) |> xfun::md_table()

Error in match.arg(digits) : 'arg' must be NULL or a character vector

5: stop("'arg' must be NULL or a character vector")
4: match.arg(digits)
3: setattr(switch(match.arg(digits), hours = as.integer(round(unclass(x)/3600) * 
       3600), minutes = as.integer(round(unclass(x)/60) * 60)), 
       "class", "ITime")
2: round.ITime(structure(86280L, class = "ITime"), 3)
1: xfun::md_table(data.frame(x = data.table::as.ITime("23:58:00")))

This happens when md_table gives 3 as the default argument to round(column): https://github.com/yihui/xfun/blob/046d83997ccd44634210d8af227019111e780b49/R/markdown.R#L293-L299

round(Sys.time(), 3) would also fail, but md_table skips columns that aren't is.numeric(). Should as.numeric.ITime return FALSE as well?