perf[gpu]: generate FixedSizeList offsets on device (#8458)

Use the existing CUDA sequence kernel to materialize Arrow List offsets
for FixedSizeList export directly on the device.

This avoids building the offsets buffer on the CPU and copying it to
CUDA memory while preserving existing i32 overflow checks.

Signed-off-by: "Alexander Droste" <alexander.droste@protonmail.com>

Author: 0ax1

vortex-cuda/src/arrow/canonical.rs

@@ -1011,7 +1011,7 @@ async fn export_fixed_size_list(
     } = array.into_data_parts();
 
     let (validity_buffer, null_count) = export_arrow_validity_buffer(validity, len, 0, ctx).await?;
-    let offsets_buffer = fixed_size_list_offsets(len, list_size, ctx).await?;
+    let offsets_buffer = fixed_size_list_offsets(len, list_size, ctx)?;
 
     export_list_layout(
         elements,
@@ -1025,7 +1025,7 @@ async fn export_fixed_size_list(
     .await
 }
 
-async fn fixed_size_list_offsets(
+fn fixed_size_list_offsets(
     len: usize,
     list_size: u32,
     ctx: &mut CudaExecutionCtx,
@@ -1035,17 +1035,29 @@ async fn fixed_size_list_offsets(
             "cannot export FixedSizeList with list size {list_size}: Arrow List offsets require i32"
         )
     })?;
-    let offsets = (0..=i32::try_from(len)?)
-        .map(|idx| {
-            idx.checked_mul(list_size)
-                .ok_or_else(|| vortex_err!("FixedSizeList Arrow List offsets exceed i32 range"))
-        })
-        .collect::<VortexResult<Vec<_>>>()?;
+    let len_i32 = i32::try_from(len)?;
+    len_i32
+        .checked_mul(list_size)
+        .ok_or_else(|| vortex_err!("FixedSizeList Arrow List offsets exceed i32 range"))?;
+
+    let output_len = len
+        .checked_add(1)
+        .ok_or_else(|| vortex_err!("FixedSizeList Arrow List offsets length overflows usize"))?;
+    let offsets = ctx.device_alloc::<i32>(output_len)?;
+    let base = 0i32;
+    let output_len_u64 = output_len as u64;
+    let kernel = ctx.load_function_with_suffixes("sequence", &["i32"])?;
+
+    ctx.launch_kernel(&kernel, output_len, |args| {
+        args.arg(&offsets)
+            .arg(&base)
+            .arg(&list_size)
+            .arg(&output_len_u64);
+    })?;
 
-    ctx.ensure_on_device(BufferHandle::new_host(
-        Buffer::from(offsets).into_byte_buffer(),
-    ))
-    .await
+    Ok(BufferHandle::new_device(Arc::new(CudaDeviceBuffer::new(
+        offsets,
+    ))))
 }
 
 /// Return Arrow Device `List` offsets as an `i32` device buffer.