feat: search multiple query vectors as one query (#1811)

Allows users to pass multiple query vector as part of a single query
plan. This just runs the queries in parallel without any further
optimization. It's mostly a convenience.

Previously, I think this was only handled by the sync Python remote API.
This makes it common across all SDKs.

Closes https://github.com/lancedb/lancedb/issues/1803

```python
>>> import lancedb
>>> import asyncio
>>> 
>>> async def main():
...     db = await lancedb.connect_async("./demo")
...     table = await db.create_table("demo", [{"id": 1, "vector": [1, 2, 3]}, {"id": 2, "vector": [4, 5, 6]}], mode="overwrite")
...     return await table.query().nearest_to([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [4.0, 5.0, 6.0]]).limit(1).to_pandas()
... 
>>> asyncio.run(main())
   query_index  id           vector  _distance
0            2   2  [4.0, 5.0, 6.0]        0.0
1            1   2  [4.0, 5.0, 6.0]        0.0
2            0   1  [1.0, 2.0, 3.0]        0.0
```

nodejs/__test__/table.test.ts

@@ -998,4 +998,18 @@ describe("column name options", () => {
     const results = await table.query().where("`camelCase` = 1").toArray();
     expect(results[0].camelCase).toBe(1);
   });
+
+  test("can make multiple vector queries in one go", async () => {
+    const results = await table
+      .query()
+      .nearestTo([0.1, 0.2])
+      .addQueryVector([0.1, 0.2])
+      .limit(1)
+      .toArray();
+    console.log(results);
+    expect(results.length).toBe(2);
+    results.sort((a, b) => a.query_index - b.query_index);
+    expect(results[0].query_index).toBe(0);
+    expect(results[1].query_index).toBe(1);
+  });
 });

nodejs/lancedb/query.ts

@@ -492,6 +492,42 @@ export class VectorQuery extends QueryBase<NativeVectorQuery> {
     super.doCall((inner) => inner.bypassVectorIndex());
     return this;
   }
+
+  /*
+   * Add a query vector to the search
+   *
+   * This method can be called multiple times to add multiple query vectors
+   * to the search. If multiple query vectors are added, then they will be searched
+   * in parallel, and the results will be concatenated. A column called `query_index`
+   * will be added to indicate the index of the query vector that produced the result.
+   *
+   * Performance wise, this is equivalent to running multiple queries concurrently.
+   */
+  addQueryVector(vector: IntoVector): VectorQuery {
+    if (vector instanceof Promise) {
+      const res = (async () => {
+        try {
+          const v = await vector;
+          const arr = Float32Array.from(v);
+          //
+          // biome-ignore lint/suspicious/noExplicitAny: we need to get the `inner`, but js has no package scoping
+          const value: any = this.addQueryVector(arr);
+          const inner = value.inner as
+            | NativeVectorQuery
+            | Promise<NativeVectorQuery>;
+          return inner;
+        } catch (e) {
+          return Promise.reject(e);
+        }
+      })();
+      return new VectorQuery(res);
+    } else {
+      super.doCall((inner) => {
+        inner.addQueryVector(Float32Array.from(vector));
+      });
+      return this;
+    }
+  }
 }
 
 /** A builder for LanceDB queries. */

nodejs/src/query.rs

@@ -135,6 +135,16 @@ impl VectorQuery {
         self.inner = self.inner.clone().column(&column);
     }
 
+    #[napi]
+    pub fn add_query_vector(&mut self, vector: Float32Array) -> Result<()> {
+        self.inner = self
+            .inner
+            .clone()
+            .add_query_vector(vector.as_ref())
+            .default_error()?;
+        Ok(())
+    }
+
     #[napi]
     pub fn distance_type(&mut self, distance_type: String) -> napi::Result<()> {
         let distance_type = parse_distance_type(distance_type)?;