pa.Table has different length than its column

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

If I run the following script

```python
import pyarrow as pa
import numpy as np

tables = [
pa.table({"x": [1, 2, 3]}),
pa.table({"x": [4]}),
pa.table({"x": [5, 6]}),
]

sa = pa.concat_tables(tables).to_struct_array().combine_chunks()

offsets = [0, 0, 0, 0]
offsets[1:] = np.cumsum([len(t) for t in tables])

la = pa.ListArray.from_arrays(offsets=offsets, values=sa)

for i in range(3):
value = la[i].values
table = pa.Table.from_struct_array(value)
print(len(value), len(table), len(table["x"]))
```

I get the following output:

```
3 3 6
1 1 1
2 2 2
```

This means that for some reason, table has a different length than one of its constituent columns, which should be impossible!

I am running this using Python 3.12 + PyArrow 21 (installed via conda-forge on Fedora Linux).
### Component(s)

Python