Function execute_tiled 

pub fn execute_tiled<F>(
    raster: &[f32],
    width: usize,
    height: usize,
    config: &TiledConfig,
    tile_fn: F,
) -> GpuResult<Vec<f32>>
where
    F: Fn(&RasterTile) -> GpuResult<Vec<f32>>,

High-level tiled executor: split → per-tile tile_fn → stitch.

tile_fn(tile) -> GpuResult<Vec<f32>> receives each tile and must return a Vec<f32> whose length equals tile.padded_len() (the same layout as tile.data). GPU submission typically happens inside tile_fn.

The stitcher discards the halo and writes only the core interior into the full-resolution output.

Errors

Returns the first error propagated by tile_fn.