Function logsumexp_last_dim 

pub fn logsumexp_last_dim(input: &Tensor) -> Result<Tensor, KernelError>

Log-sum-exp reduction along the last tensor dimension.

Returns shape equal to input with the last dimension set to 1.