Added ability to inspect a 'Sequence' decoder and the AddedVocabulary.

[eaplatanios]

This PR is in similar spirit to #1341 and adds a couple more functions that allow one to construct a modified version of an existing Tokenizer. I've followed the existing style and conventions for newly introduced functions.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[eaplatanios]

@ArthurZucker @Narsil gentle ping on this one and on #1444.

[eaplatanios]

@ArthurZucker @Narsil gentle ping about this PR. This PR should not be controversial and is in similar spirit to #1341 (and has the same motivation).

[ArthurZucker]

really sorry about all the delays, lot happening on transformers, I'll free some time

[ArthurZucker]

Actually doesn't look bad and would help me in transformers if we add the layer to update the added vocab. This is not super stateless, so leave it up to @Narsil

[ArthurZucker]

Also can you add tests for set and get? 🤗

[Narsil]

LGTM.

It's ok because it doesn´t seem like we open up any function that have potentially bad internal state (Deserialize is not implemented for AddedVocabulary directly, itś always parsed from the entire Tokenizer).

[eaplatanios]

Sorry this fell through the cracks a bit over the past couple of weeks and I just saw the last couple of comments. Thanks for approving and merging this!

[ArthurZucker]

Thanks for your contribution 🤗