Add MLX GEMM dispatch

Description

mlx has a GEMM function that they call mlx.core.addmm. We can dispatch our GEMM Op to it as follows:

@mlx_funcify.register(Gemm)
def mlx_funcify_Gemm(op, **kwargs):
    # GEMM has signature:
    # b * z + a * dot(x, y)
    
    def gemm(z, a, x, y, b):
        # mx.addmm has signature:
        # alpha * (a @ b)  + beta * c    
        return mx.addmm(z, x, y, alpha=a, beta=b)
    
    return gemm

What's tricky is that the blas rewrite machinery is quite C specific. I'm not sure if we can register just the dot22_to_gemm rewrite for sure in MLX.

Start by reading the existing mlx_funcify registrations and the Gemm entry point shown in the issue, then trace the dot22_to_gemm rewrite and its BLAS-specific assumptions. Confirm how MLX exposes addmm and whether that rewrite can be registered independently; done means GEMM dispatches through addmm and the applicable rewrite path works without C-specific machinery.