Rampy

- Removing dependency to cvxpy that does not build well in Windows... It affects the use of the rampy.mixing function

- Correction of an error in `mlregressor` which made impossible to import X_test datasets.

- BREAKING CHANGE: `mlregressor` is now a class and not a function anymore. You can provide directly sklearn arguments to the algorithms through dictionaries.
The use of the class simplifies the use of `mlregressor`, as the created objects saves everything!
It also makes it very easy to change the algorithm and try something else. See the example in the example folder!

- addition of the `centroid()` function, that calculates the centroid of a signal.

- addition of tests and examples for the `mlregressor()` class, the `resample()` and `flipsp()` functions.

- `chemical_splitting()` allows one to select the random seed.

- Correction of the `mixing_sp()` function, rampy is now compatible with cvxpy v1.0.

- arguments can be provided to `resample()` to use different techniques of interpolation in `scipy.interpolate.interp1d`.

- Various documentation improvements

- Correction of the `tlcorrection()` function: the 'hehlen' correction was missing a frequency term to be complete (eq. 2 and 3 in Hehlen 2010 J. Phys. Condes. Matter 22: 025401).

- Addition of the `rampy.mixing_sp()` function. See `help(rampy.mixing_sp())`, as well as the example folder.

- gcvspline is not a requirement anymore. Error messages will outputs when trying to use it, inviting to install it manually. This is implemented to avoid problems with FORTRAN compilation for people not interested in using gcvspline.

- Add early stopping in` mlregressor` neural networks.