Function matmul 

pub fn matmul(
    a: &Tensor,
    b: &Tensor,
    bias: Option<&Tensor>,
    spec: MatmulSpec,
) -> Result<Tensor>

Batched matmul with optional A/B-transpose, scaling, and bias.

A: [m, k] or [batch, m, k] ([k, m] with trans_a); B: [k, n] ([n, k] transposed), optionally batched. A rank-2 operand broadcasts across the batch. Bias: [n].