[Python][Parquet] Cannot read parquet with duplicate column names

### Describe the bug, including details regarding any error messages, version, and platform.

If `use_legacy_dataset=False` in `use_legacy_dataset` then Exception is raised.
If we specify the `schema`, the read succeeds, but the values of the duplicate columns are `null`.

Using `use_legacy_dataset=True` there are no issues reading the single file (but we can only `rename_columns` after reading the whole file).

Alternatively the `columns` param of `read_table` should allow reading using index, not just column names.

### Component(s)

Parquet, Python