[BUG]: ordering produces extremely high reads and low performance.

### What version of `drizzle-orm` are you using?

0.29.0

### What version of `drizzle-kit` are you using?

_No response_

### Describe the Bug

with the following schemas:
```javascript
export const debatesTable = mysqlTable(
"debates",
{
id: v4IDWithDefault("id").primaryKey(),
name: varchar("name", { length: 255 }).unique().notNull(),
},
);
export const debatesRelations = relations(debatesTable, ({ one, many }) => ({
posts: many(postsTable),
}));

export const postsTable = mysqlTable(
"posts",
{
id: v4IDWithDefault("id").primaryKey(),
authorId: v4Id("author_id").notNull(),
content: text("content").notNull(),
debateId: v4Id("debate_id").notNull(),
createdAt: timestamp("created_at").defaultNow().notNull(),
likeCount: mediumint("like_count").default(0).notNull(),
},
(table) => {
return {
debateIdIdx: index("debate_id_idx").on(table.debateId),
// this is DESC but added manually because drizzle does not support it yet
createdAtIdx: index("created_at_idx").on(table.createdAt),
createdAtDebateIdIdx: index("created_at_debate_id_idx").on(
table.debateId,
table.createdAt
),
};
}
);

export const postsRelations = relations(postsTable, ({ one, many }) => ({
author: one(usersTable, {
fields: [postsTable.authorId],
references: [usersTable.id],
}),
debate: one(debatesTable, {
fields: [postsTable.debateId],
references: [debatesTable.id],
}),
}));
```

The problem happens when running the following query:
```javascript
db.query.debatesTable.findFirst({
where: (debate, { eq }) => eq(debate.name, debateName),
with: {
posts: {
orderBy: desc(postsTable.createdAt),
limit: 16,
},
},
});
```

among other things, there's a part in the raw SQL that looks like this:
```sql
FROM
(SELECT *, row_number() OVER (
ORDER BY `debatesTable_posts`.`created_at` DESC)
FROM `posts` `debatesTable_posts`
WHERE `debatesTable_posts`.`debate_id` = `debatesTable`.`id`
LIMIT 16) `debatesTable_posts`), json_array()) AS `posts`
```

I faced extremely high row reads and I contacted planetscale support, and the person helping me pointed out that calling `row_number()` seemed to make the engine look into every row. I tried to run a modified version of the query that looked like this:

```sql
FROM
(SELECT *
FROM `posts` `debatesTable_posts`
WHERE `debatesTable_posts`.`debate_id` = `debatesTable`.`id`
LIMIT 16 ORDER BY `debatesTable_posts`.`created_at` DESC) `debatesTable_posts`), json_array()) AS `posts`
```
and I got around 37 row reads, as opposed to the 21.000 I was getting before.

### Expected behavior

Less row reads

### Environment & setup

Planetscale, drizzle 0.29.0