Respect row ordering when concatenating datasets along axis=1

[mariosasko]

Currently, there is a bug when concatenating datasets along axis=1 if more than one dataset has the _indices attribute defined. In that scenario, all indices mappings except the first one get ignored.

A minimal reproducible example:

>>> from datasets import Dataset, concatenate_datasets
>>> a = Dataset.from_dict({"a": [30, 20, 10]})
>>> b = Dataset.from_dict({"b": [2, 1, 3]})
>>> d = concatenate_datasets([a.sort("a"), b.sort("b")], axis=1)
>>> print(d[:3])              # expected: {'a': [10, 20, 30], 'b': [1, 2, 3]}
{'a': [10, 20, 30], 'b': [3, 1, 2]}

I've noticed the bug while working on #3195.