Support composite types with fields that are of type `citext`

Nobody has claimed this yet.

* **asyncpg version**: 0.28
* **PostgreSQL version**: 15
* **Do you use a PostgreSQL SaaS? If so, which? Can you reproduce
the issue with a local PostgreSQL install?**: Local version
* **Python version**: 3.11
* **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?**: -
* **Can the issue be reproduced under both asyncio and
[uvloop](https://github.com/magicstack/uvloop)?**: not tested

I am using citext for fields that really need to be citext (as in, I care about the case only for displaying but other than that the case should not matter). Now I have a query that goes like

```sql
SELECT * FROM project WHERE (owner::citext, name::citext)::record = ANY($1);
```

Throughout asyncpg it is mentioned to get this to work (normally) you add a composite type. So we add a composite type and cast the array to the composite type.
```sql
CREATE TYPE project_fqn AS (
owner citext,
name citext
);

SELECT * FROM project WHERE (owner::citext, name::citext) = ANY($1::project_fqn[]);
```

Sadly this fails due to an `: cannot decode type "public"."project_fqn": text encoding of composite types is not supported.` error. It can be "fixed" by changing all `citext` types for a `text` type but then we lose the case insensitive search.

Is this something that can be fixed or are there any workarounds? I tried doing a `set_builtin_type_codec("citext", codec_name=25)` but that seems to have no impact.

Some references:
- https://github.com/MagicStack/asyncpg/issues/476
- https://github.com/MagicStack/asyncpg/issues/82

Reproduce the reported SQL using asyncpg 0.28 with PostgreSQL 15 and a composite type whose fields are citext. Start by tracing composite decoding and the set_builtin_type_codec("citext") entry point; done means project_fqn[] values decode without UnsupportedClientFeatureError while preserving citext behavior.

Written by the indexing model from the issue text.