pub const MOE_ROW_UNION_MAX_IDS: usize = 64;
2026-09-25: Most ids glm5next_moe_row_union resolves: it runs as one block of rows * top_k threads, and forward_moe does not take the row-batched path above this.
glm5next_moe_row_union
rows * top_k
forward_moe