`information_schema.columns` returns blank metadata instead of view columns

Doltgres does not expose view output columns correctly through `information_schema.columns`. Each view is represented by a single fabricated row with an empty column name, ordinal position `0`, and no type metadata.

### Reproduction

```sql
CREATE TABLE view_metadata_probe (
id integer,
label varchar(20)
);

CREATE VIEW view_metadata_probe_v AS
SELECT id, label
FROM view_metadata_probe;

SELECT
column_name,
ordinal_position,
data_type,
udt_schema,
udt_name
FROM information_schema.columns
WHERE table_schema = 'public'
AND table_name = 'view_metadata_probe_v'
ORDER BY ordinal_position;
```

### Doltgres result

The view is represented by one row resembling:

```text
column_name | ordinal_position | data_type | udt_schema | udt_name
------------+------------------+-----------+------------+---------
| 0 | NULL | NULL | NULL
```

### PostgreSQL result

PostgreSQL 16 returns:

```text
column_name | ordinal_position | data_type | udt_schema | udt_name
------------+------------------+-------------------+------------+---------
id | 1 | integer | pg_catalog | int4
label | 2 | character varying | pg_catalog | varchar
```

### Impact

Schema-discovery tools and database clients cannot determine a view's columns or types through the standard information schema, even though the underlying view data is available.

### Implementation context

`getRowsFromViews` in `server/tables/information_schema/columns_table.go` currently appends one literal placeholder row for each view. It does not analyze the view definition or iterate over its output schema.

The `pg_attribute` catalog already analyzes view queries and emits one entry per output column. The information-schema implementation should similarly derive the analyzed view schema and build a normal metadata row for each output column.

This was identified as a pre-existing issue while validating dolthub/doltgresql#3241.

Start in server/tables/information_schema/columns_table.go at getRowsFromViews and run the SQL reproduction against a view. Compare its placeholder-row logic with the pg_attribute catalog's analyzed view schema handling. Done means information_schema.columns returns one row per view output column with the correct name, ordinal position, and type metadata.

Written by the indexing model from the issue text.