[Feature request] Initialise with covariance matrix

[jpiabrantes]

Currently we can only initialise the solver with a scalar sigma, or (maybe) with diagonal matrix using the options?

I want to initialise using a full covariance matrix.

[nikohansen]

See #236 and #255.

Generally, changing the initial covariance matrix from the identity to $C_0$ is equivalent with applying the linear transformation $\sqrt{C_0}^{-1}$ on the search space before calling the objective functions and the latter is generally preferable to the former. The latter can be implemented as a simple wrapper around the objective function. An implementation of the respective transformation which would be equivalent with setting the respective initial sample covariance matrix, or, in other words, "effectively change the covariance matrix without changing the covariance matrix" can be found here.