[Python] Define whether user-overridden to_pylist on Array subclasses is honored for nested children

In the review of #50327 it was noted (https://github.com/apache/arrow/pull/50327#discussion_r3629728508) that the bulk conversion does not consult a user-defined `to_pylist` override on `ExtensionArray` subclasses for *nested* children.

This was already inconsistent before #50327:
- `ListScalar.as_py` called `child_values.to_pylist()`, so an override on a list child's class was honored;
- `StructScalar.as_py` converted fields via per-element Scalars, so an override on a struct child's class was ignored.

Top-level overrides dispatch normally through Python method resolution in both the old and the new code, and `ExtensionScalar.as_py` / `__arrow_ext_scalar_class__` — the documented customization point — is honored on all paths.

This issue is to decide the contract:
1. document that nested `to_pylist` overrides are not consulted (only `ExtensionScalar.as_py` is), or
2. detect custom `to_pylist` overrides and route nested conversion through them (sketch in the linked discussion).

Start with the linked discussion in #50327 and compare the existing ListScalar.as_py and StructScalar.as_py paths. Decide whether nested to_pylist overrides are part of the contract or whether only ExtensionScalar.as_py is supported; done means the chosen behavior is consistently implemented and documented, with coverage for nested children.

Written by the indexing model from the issue text.