add `tokenize_document` from `pie-modules`

`tokenize_document` is PyTorch specific (it uses a `transformers` tokenizer), so it will be removed from `pie-modules` (see https://github.com/ArneBinder/pie-modules/pull/187).

more context: https://github.com/ArneBinder/pie-core/issues/17

requires `pie-modules` release including https://github.com/ArneBinder/pie-modules/pull/187 