hf_load_dataset() shouldn't default to using labels

[samterfa]

Some datasets don't have labels defined but are still valid datasets (one I'm using for the ez functions vignette). hf_load_dataset("sentiment140", split = "test") results in KeyError: 'label'.

[jpcompartir]

These are the offending blocks - we can re-implement fairly straightforwardly to cover this dataset, but would like to implement something more general so thought required.

#Set the default value of label_conversion to 'int2str' unless specified, in which case match the input
  label_conversion <- match.arg(if (missing(label_conversion)) "int2str" else label_conversion, c("str2int", "int2str"))

 #get int2str & str2int which can later be called directly on the label variable
  if(!is.null(label_conversion)){
    x <- splits[[1]]
    x <- .dataset[[x]]
    x <- x[["features"]]
    x <- x[["label"]]
    int2str <- x[["int2str"]]
    str2int <- x[["str2int"]]
  }

  if(label_conversion == "int2str"){
    label_names <- purrr::map(datasets, ~int2str(.x[["label"]]))
    datasets <- purrr::map2(.x = datasets, .y = label_names, .f = ~ .x %>% dplyr::mutate(label_name = .y))
  }
  if(label_conversion == "str2int"){
    label_ids <- purrr::map(datasets, ~str2int(.x[["label"]]))
    datasets <- purrr::map2(.x = datasets, .y = label_ids, .f = ~ .x %>% dplyr::mutate(label_id = .y))
  }