please implement the from_pretrained tokenizer method

[cregouby]

Hello,

current behavior

The Quicktour from Huggingface ends with the following paragraph :

but currently, {hftokenizer} do not allow to load pretrained tokenizers.

expected behavior

I'd like to be able to reuse pretrained tokenizers already available in the LM models present in the wild ( BERT, RoBERTa and friends) and / or in my local cache folder, in order to feed those models with the result of {hftokenizer} tokenizer$encode()$ids.

And have the Quicktour vignette to cover the API to do it.

Thanks a lot !