fix(python): normalize blob row ID inputs and errors (#4293)

nodejs/__test__/table.test.ts

@@ -2817,6 +2817,18 @@ describe("when dealing with blob columns", () => {
     expect(Buffer.from(await files[1]!.read()).toString()).toBe("beta");
   });
 
+  it("rejects a whole blob read batch when one row ID was deleted", async () => {
+    const { table, rowIds } = await openBlobTable();
+    await table.delete("id = 2");
+
+    await expect(table.fetchBlobs("image", rowIds)).rejects.toThrow(
+      `first missing row ids: [${rowIds[1]}]`,
+    );
+    await expect(table.fetchBlobFiles("image", rowIds)).rejects.toThrow(
+      `first missing row ids: [${rowIds[1]}]`,
+    );
+  });
+
   it("adds a LargeBinary Arrow column to a blob table", async () => {
     const db = await connect(tmpDir.name);
     const schema = new Schema([new Field("id", new Int64()), blob("image")]);

nodejs/lancedb/table.ts

@@ -536,6 +536,8 @@ export abstract class Table {
    * Reads the table's current checkout. IDs from another version can fail after
    * compaction unless stable row ids are enabled. Results keep input order and
    * duplicates. Null blobs are `null`. Empty blobs are empty buffers.
+   * One missing or deleted row ID rejects the entire batch, including on remote
+   * tables; no partial results are returned.
    * Remote servers limit each request to 1024 row IDs and 64 MiB of blob bytes.
    * The client splits requests automatically and reads an individual larger
    * blob through the Range route. This method still materializes all bytes in
@@ -553,6 +555,8 @@ export abstract class Table {
    *
    * Preserves input order, duplicates, and nulls. Use this for large payloads.
    * See {@link Table.fetchBlobs} for row-ID validity across versions.
+   * One missing or deleted row ID rejects the entire batch, including on remote
+   * tables; no partial handles are returned.
    */
   abstract fetchBlobFiles(
     column: string,