[Python] [Parquet] support nested column selection for list-struct types when reading parquet files

### Describe the enhancement requested

Currently, `pyarrow.parquet.read_table(columns)` supports selection of nested columns with dot notation. For example, `columns=["a.b"]` will select field "b" from the struct column "a". It would be really nice if this also worked for list columns with struct values.

Currently, this code fails:

```python
import pyarrow as pa
import pyarrow.parquet as pq

list_struct = pa.ListArray.from_arrays(
values = [{"b": 1, "c": 1}] * 10,
offsets = [0, 4, 10],
)
pq.write_table(pa.table({"a": list_struct}), "/tmp/table.parquet")
pq.read_table("/tmp/table.parquet", columns=["a.b"])
```

Traceback

```
File ~/.virtualenvs/v/lib/python3.13/site-packages/pyarrow/parquet/core.py:1824, in read_table(source, columns, use_threads, schema, use_pandas_metadata, read_dictionary, memory_map, buffer_size, partitioning, filesystem, filters, ignore_prefixes, pre_buffer, coerce_int96_timestamp_unit, decryption_properties, thrift_string_size_limit, thrift_container_size_limit, page_checksum_verification)
1812 # TODO test that source is not a directory or a list
1813 dataset = ParquetFile(
1814 source, read_dictionary=read_dictionary,
1815 memory_map=memory_map, buffer_size=buffer_size,
(...) 1821 page_checksum_verification=page_checksum_verification,
1822 )
-> 1824 return dataset.read(columns=columns, use_threads=use_threads,
1825 use_pandas_metadata=use_pandas_metadata)

File ~/.virtualenvs/v/lib/python3.13/site-packages/pyarrow/parquet/core.py:1475, in ParquetDataset.read(self, columns, use_threads, use_pandas_metadata)
1467 index_columns = [
1468 col for col in _get_pandas_index_columns(metadata)
1469 if not isinstance(col, dict)
1470 ]
1471 columns = (
1472 list(columns) + list(set(index_columns) - set(columns))
1473 )
-> 1475 table = self._dataset.to_table(
1476 columns=columns, filter=self._filter_expression,
1477 use_threads=use_threads
1478 )
1480 # if use_pandas_metadata, restore the pandas metadata (which gets
1481 # lost if doing a specific `columns` selection in to_table)
1482 if use_pandas_metadata:

File ~/.virtualenvs/v/lib/python3.13/site-packages/pyarrow/_dataset.pyx:579, in pyarrow._dataset.Dataset.to_table()

File ~/.virtualenvs/v/lib/python3.13/site-packages/pyarrow/_dataset.pyx:415, in pyarrow._dataset.Dataset.scanner()

File ~/.virtualenvs/v/lib/python3.13/site-packages/pyarrow/_dataset.pyx:3676, in pyarrow._dataset.Scanner.from_dataset()

File ~/.virtualenvs/v/lib/python3.13/site-packages/pyarrow/_dataset.pyx:3589, in pyarrow._dataset.Scanner._make_scan_options()

File ~/.virtualenvs/v/lib/python3.13/site-packages/pyarrow/_dataset.pyx:3539, in pyarrow._dataset._populate_builder()

File ~/.virtualenvs/v/lib/python3.13/site-packages/pyarrow/error.pxi:92, in pyarrow.lib.check_status()

ArrowInvalid: No match for FieldRef.Nested(FieldRef.Name(a) FieldRef.Name(b)) in a: list>
__fragment_index: int32
__batch_index: int32
__last_in_fragment: bool
__filename: string
```

**Edit 2025.05.06**: add context:

In our projects ([HATS](https://github.com/astronomy-commons/hats), [LSDB](https://github.com/astronomy-commons/lsdb), [nested-pandas](https://github.com/lincc-frameworks/nested-pandas)), we use nested structures intensively, especially list-structured arrays/columns. Unfortunately, we haven't found a simple way to load these columns partially, for the sake of minimizing I/O and memory footprint. As a result, we have to store the data as struct-columns, which introduces the overhead of storing additional offsets and also doesn't provide a simple way to guarantee that all offsets are identical.

### Component(s)

Python