[Feature] Diagonal operations handling in HamiltonianEvolution

[chMoussa]

Related to issue #211. This is an experimental feature to be checked with Qadence.

For generators that contains only diagonal operations, we avoid computing the whole tensor and use only the resulting diagonal. For this we introduce the option diagonal in certain operations. If set to True, the returned tensor would be of shape (2^n, batch_dim) if the operation is really diagonal. One can convert to the diagonal form using the to_diagonal method. Then apply methods have a new attribute that use different einsum indices for such cases.

At the moment, we will only support such diagonal operations in HamiltonianEvolution, for the generator only. The returned tensors from HamiltonianEvolution would still be of shape (2^n, 2^n, batch_dim) to be compatible with Qadence.