Nested custom types in Postgres break `asyncpg.Record` subclassing

* **asyncpg version**: v0.25.0

* **PostgreSQL version**: 14

* **Do you use a PostgreSQL SaaS? If so, which? Can you reproduce
the issue with a local PostgreSQL install?**: on-premise

* **Python version**: 3.10.1
* **Platform**: linux
* **Do you use pgbouncer?**: no
* **Did you install asyncpg with pip?**: yes
* **If you built asyncpg locally, which version of Cython did you use?**: n/a
* **Can the issue be reproduced under both asyncio and
[uvloop](https://github.com/magicstack/uvloop)?**: n/a

It would appear that nested types are not converted to a user specified `record_class`, and instead will always use `asyncpg.Record`.

For example:

```

CREATE TYPE my_custom_type AS (
id int,
filename text,
sql text
);

CREATE TABLE my_table (
files my_custom_type[] NOT NULL,
);

SELECT * FROM my_table
```

If you specify a custom `record_class`, then the record from `SELECT * FROM my_table` will be your subclass. However, the value inside `record['files']` will use the `asyncpg.Record` class, rather than the custom one specified.

## What should actually happen?

I think if a user specifies a `record_class`, then this class should always be used, even for nested custom types. It's quite unexpected behaviour for `asyncpg` to use the default class when a custom one has been specified.

Start by reproducing the issue with the custom PostgreSQL type, array column, SELECT query, and a custom record_class described in the report. Trace how nested custom-type values are decoded and how record_class is propagated; done means records inside record['files'] use the specified subclass rather than asyncpg.Record, with regression coverage for this case.

Written by the indexing model from the issue text.