Unexpected/undesirable `CAST` of date string to `VARCHAR`

Nobody has claimed this yet.

* **asyncpg version**: 0.29.0
* **PostgreSQL version**: 15.7
* **Python version**: 3.11
* **Platform**: macOS 13.6
* **Do you use pgbouncer?**: no
* **Did you install asyncpg with pip?**: no (poetry)

```python
start_date = '2024-01-01'
query.where(MyTable.datetime_col >= start_date)
```

This will fail:

```
: operator does not exist: timestamp without time zone >= character varying
```

It seems that this is being casted as `VARCHAR`:

```
SELECT * FROM my_table WHERE datetime_col >= $1::VARCHAR
```

This same filter is valid in Postgres

```sql
SELECT * FROM my_table WHERE datetime_col >= '2024-01-01'
```

It works when using the datetime object:

```python
start_date = dt.datetime.strptime(start_date, "%Y-%m-%d")
query.where(MyTable.datetime_col >= start_date)
```

Wonder if this is somewhat similar to #1169 in the sense that casting/argument handling invalidates valid SQL statements.

I'd have imagined that castings were performed in obvious and non-breaking scenarios, and scenarios where casting would be necessary, but are not obvious should be handled directly by the user. Breaking valid SQL statements seems counter intuitive IMHO.

Reproduce the query.where case using asyncpg 0.29.0, PostgreSQL 15.7, and Python 3.11, comparing the string and datetime inputs. Inspect the generated SQL and parameter handling around the reported $1::VARCHAR cast. Done means the expected date comparison behavior is established and covered by a regression test or the issue is shown to belong elsewhere.

Written by the indexing model from the issue text.