Initialize eigh by previous iteration.

[AlexanderMath]

Every iteration of DFT computes an eigendecomposition at (line 49). Since we compute the eigendecomposition at every iteration, we may be able to initialize the eigensolver at iteration i by the eigenvector solution from iteration i-1.

Notes:

1. The eigensolver is implemented in tesselate.
2. tesselate has two eigensolvers: the jacobi algorithm and the QR algorithm.
3. The matrix may change a lot iteration 0 to iteration 1 and less onwards.

[awf]

Looks like we would add a all_AV_cols argument to

def ipu_jacobi_eigh(x: Array, num_iters: int = 1) -> Tuple[Array, Array]: