Function k1_ssd_chunk_cumsum 

pub fn k1_ssd_chunk_cumsum(
    dt_discretized_bhnl: Tensor<4>,
    a_decay_h: Tensor<1>,
) -> (Tensor<4>, Tensor<3>)

Based on the Kernel 1 Triton reference _chunk_cumsum_fwd_kernel (ssd_chunk_state.py).

Returns:

• da_cumsum_bhnl [used in K3+K5][*] - intra-chunk cumsum.
• da_chunk_end_bhn [used in K4][omitted][*] - last da_cumsum per chunk.