update to_py_obj to support np.number

[PrettyMeng]

What does this PR do?

Update to_py_obj function to make it support np.number type.

Fixes # See an example below:

tokenizer.decode(110)
'no'
tokenizer.decode(np.array([110])[0])
Traceback (most recent call last):
File "", line 1, in
File "/Users/prettymeng/anaconda3/envs/ptuning/lib/python3.8/site-packages/transformers/tokenization_utils_base.py", line 3179, in decode
return self._decode(
File "/Users/prettymeng/anaconda3/envs/ptuning/lib/python3.8/site-packages/transformers/tokenization_utils_fast.py", line 527, in _decode
text = self._tokenizer.decode(token_ids, skip_special_tokens=skip_special_tokens)
TypeError: Can't convert 110 to Sequence

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[PrettyMeng]

@sgugger Hi! This is an reopened PR for #14041, could you please check whether the failed tests are related to my PR?

[sgugger]

All good, thanks a lot for your contribution!