added history to mackey_glass

[arashgmn]

Currently, delayed differential equations (DDE) such as Mackey-Glass are initialized with random history. Yet, for certain tasks, it is desirable to directly set the history. Consider estimating the Lyapunov exponent, for instance. One may want to generate two trajectories with initial conditions $d_0$ apart. The initial condition of DDE, however, also includes the historical record of the process. Thus, a correct setup must use the following definition of distance:

$\textbf d^2_{ij} = \frac{1}{\tau}\int_ {t=-\tau}^0 || \textbf{x}_i(t) - \textbf{x}_j(t) ||^2 dt$

which in discrete case translates into:

$\textbf d^2_{ij} = \Big[ \delta x[0]^2 + ( \delta x[-1]^2 + \delta x[-2]^2 + ... + \delta x[-\tau/dt]^2 ) \Big] $

This PR provides the possibility to directly provide the historical sequence to the mackey_glass process via an argument history. It asserts the length of the provided array. If it is long enough, enqueues the most recent elements to the history queue, otherwise and error will be raised. The default is set to None which initializes history randomly, as usual.

[nTrouvain]

Hello @arashgmn,
Thank you so much for your contribution! We would really like to integrate it to reservoirpy.
I just made a small comment on the nature and content of the check you perform on the history variable. Also, if it is possible for you, could you add some elements to the datasets unit tests to check that your check passes correctly, and maintain coverage over this function?

[nTrouvain]

I think this should be a ValueError. AssertionError are not specific enough. And maybe the error text is a bit puzzling: maybe you could just state that history must have a length of a specific value, and add in parenthesis that this value is equal to tau/h ?