Reconstruct chains for SamplingResult

[SarahRo]

Add load_data method to SamplingResult that reconstructs chains

[NicolaCourtier]

To resolve this issue, we should add the x, x_model, x0, cost and iteration_number data to the dictionary created in Result.save_data. I think we may be able to write a function SamplingResult.chains that returns the current chains attribute from the data stored in x_model, instead of having different attributes between the result classes. Note that chains is an array with dimensions (n_parameters, n_chains, n_iterations), not including the warm-up iterations. On the other hand, result.x_model includes all iterations and it is recorded as a list due to possible inhomogeneous list lengths due to a) optimiser runs that stop after a different number of iterations and b) a varying number of parallel evaluations per iteration returned by some optimisers.