feat(napi): Issue queries as node SDK (#868)

* Query as a fluent API and `AsyncIterator<RecordBatch>`
* Much more docs
* Add tests for auto infer vector search columns with different
dimensions.

rust/vectordb/src/ipc.rs

@@ -16,10 +16,10 @@
 
 use std::io::Cursor;
 
-use arrow_array::RecordBatchReader;
-use arrow_ipc::reader::StreamReader;
+use arrow_array::{RecordBatch, RecordBatchReader};
+use arrow_ipc::{reader::StreamReader, writer::FileWriter};
 
-use crate::Result;
+use crate::{Error, Result};
 
 /// Convert a Arrow IPC file to a batch reader
 pub fn ipc_file_to_batches(buf: Vec<u8>) -> Result<impl RecordBatchReader> {
@@ -28,6 +28,22 @@ pub fn ipc_file_to_batches(buf: Vec<u8>) -> Result<impl RecordBatchReader> {
     Ok(reader)
 }
 
+/// Convert record batches to Arrow IPC file
+pub fn batches_to_ipc_file(batches: &[RecordBatch]) -> Result<Vec<u8>> {
+    if batches.is_empty() {
+        return Err(Error::Store {
+            message: "No batches to write".to_string(),
+        });
+    }
+    let schema = batches[0].schema();
+    let mut writer = FileWriter::try_new(vec![], &schema)?;
+    for batch in batches {
+        writer.write(batch)?;
+    }
+    writer.finish()?;
+    Ok(writer.into_inner()?)
+}
+
 #[cfg(test)]
 mod tests {
 

rust/vectordb/src/query.rs

@@ -22,6 +22,7 @@ use lance_linalg::distance::MetricType;
 
 use crate::error::Result;
 use crate::utils::default_vector_column;
+use crate::Error;
 
 const DEFAULT_TOP_K: usize = 10;
 
@@ -93,6 +94,19 @@ impl Query {
                 let arrow_schema = Schema::from(self.dataset.schema());
                 default_vector_column(&arrow_schema, Some(query.len() as i32))?
             };
+            let field = self.dataset.schema().field(&column).ok_or(Error::Store {
+                message: format!("Column {} not found in dataset schema", column),
+            })?;
+            if !matches!(field.data_type(), arrow_schema::DataType::FixedSizeList(f, dim) if f.data_type().is_floating() && dim == query.len() as i32)
+            {
+                return Err(Error::Store {
+                    message: format!(
+                        "Vector column '{}' does not match the dimension of the query vector: dim={}",
+                        column,
+                        query.len(),
+                    ),
+                });
+            }
             scanner.nearest(&column, query, self.limit.unwrap_or(DEFAULT_TOP_K))?;
         } else {
             // If there is no vector query, it's ok to not have a limit