Add method to return attn-mask for HF Tokenizer.

[cptspacemanspiff]

So, I have been doing batched inference of things that use HF tokenizer. Your library is great, but does not expose the attention masks, which are useful when some of the inputs/outputs are padding.

This adds an additional tokenize method to the HF Tokenizer that returns both the token_ids and the attention masks.

It depends on my previous pull request, which separates out the hf cpp header declarations and the implementations.

#57