Function ssm_h_state_f32_to_f16 

pub fn ssm_h_state_f32_to_f16(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    src: DevicePtr,
    dst: DevicePtr,
    n: u64,
    stream: u64,
) -> Result<()>

One-shot FP32 -> FP16 conversion of one layer’s SSM h-state (ATLAS_SSM_H_FP16). n is the FP32 ELEMENT count, derived from the pool’s byte size — never a duplicated shape literal.

Kernel: ssm_h_state_f32_to_f16(src, dst, n). Grid-stride; src and dst must not alias (a narrowing compaction in place is a data race).