[C++][Python] Slicing a table with no columns returns a table with incorrect length.

Python `[:]` slicing works on tables with no columns, because the slice inputs are normalized. But the `slice` method is inconsistent.
```python

In [1]: import pyarrow as pa

In [2]: table = pa.table({'col': range(3)})

In [3]: table.slice(1).num_rows
Out[3]: 2

In [4]: table.select([])[1:].num_rows
Out[4]: 2

In [5]: table.select([]).slice(1).num_rows
Out[5]: 3

In [6]: table.select([]).slice(1, 4).num_rows
Out[6]: 4
```

**Reporter**: [A. Coady](https://issues.apache.org/jira/browse/ARROW-15412) / @coady

_{**Note**: *This issue was originally created as [ARROW-15412](https://issues.apache.org/jira/browse/ARROW-15412). Please see the [migration documentation](https://github.com/apache/arrow/issues/14542) for further details.*}

Reproduce the examples with pyarrow, comparing table.slice(1) and slicing table.select([])[1:]. Start at the C++/Python implementation behind the Table.slice entry point and trace how row counts are determined for tables with no columns. Done means slice offsets and lengths produce the same row counts as ordinary Python slicing, including the shown bounded slice.

Written by the indexing model from the issue text.