drizzle-kit push generates HNSW index DDL without operator class, causing migration failure on pgvector

## Bug

When `schema.ts` declares an HNSW index on a `vector` column with `vector_cosine_ops`, `drizzle-kit push` regenerates the index DDL without the operator class. PostgreSQL rejects the resulting `CREATE INDEX` with:

```
ERROR: no default operator class for access method "hnsw"
HINT: You must specify an operator class for the index or define a default operator class for the data type.
```

## Reproduction

`shared/schema.ts`:
```ts
import { pgTable, text, index } from "drizzle-orm/pg-core";
import { vector } from "drizzle-orm/pg-core";

export const readings = pgTable("readings", {
id: text("id").primaryKey(),
embedding: vector("embedding", { dimensions: 1536 }),
}, (table) => ({
embeddingIdx: index("idx_readings_embedding")
.using("hnsw", table.embedding.op("vector_cosine_ops")),
}));
```

Run `drizzle-kit push`. The generated DDL omits `vector_cosine_ops`:

```sql
-- Generated (broken):
CREATE INDEX "idx_readings_embedding" ON "readings" USING hnsw ("embedding");

-- Required (working):
CREATE INDEX "idx_readings_embedding" ON "readings" USING hnsw ("embedding" vector_cosine_ops);
```

## Environment

- drizzle-orm: 0.45.2
- drizzle-kit: latest as of 2026-05-21
- postgres: 16.x with pgvector 0.7+
- Node 20

## Workaround

Remove the HNSW index from `schema.ts` and manage it via raw SQL on both dev and prod:

```sql
CREATE INDEX IF NOT EXISTS idx_readings_embedding
ON readings USING hnsw (embedding vector_cosine_ops);
```

This works but defeats the purpose of declaring the index in schema.

## Expected behavior

`drizzle-kit push` should preserve operator class declarations in index definitions, especially for index methods like HNSW and IVFFlat where operator class is required.

## Related

- pgvector requires explicit operator class for HNSW/IVFFlat indexes (no default exists)
- This affects anyone using pgvector with Drizzle's `push` workflow