Function execute_reverse 

pub unsafe fn execute_reverse(
    src: usize,
    dst: usize,
    dims: &[u32],
    rev_mask: &[bool],
    elem_bytes: usize,
    base: *mut u8,
)

Host-fallback entry for Op::GatedDeltaNet (Metal / unified memory). When state == 0, uses a zero-initialized scratch state per batch item. Batch-general reverse/flip (dtype-agnostic, byte-copy). Shared by the CPU Thunk::Reverse arm and the Metal/WGPU host paths. dims is the row-major input shape; rev_mask[a] flips axis a. src/dst are byte offsets.