Refactor Spectrum class structure

[eibar-flores]

Currently PRISMA has multiple spectrum classes to deal with raw and processed spectra, see here.

Having such hierarchy of classes introduces unnecessary complexity to develop and maintain the code. I suggest sticking to a single spectrum class with three main attributes:

• index: numpy.ndarray of shape (n), holding the index of the spectrum (wavenumbers, eV, wavelenght, etc)
• counts: numpy.ndarray of shape (n), holding the counts of the spectrum (counts, counts/s etc.)
• metadata: dict with key value pairs with additional information about the spectrum. This should hold key:value pairs passed as keyword arguments when instantiating a spectrum.

Next, we should ensure all processing functions work for spectrum objects. Spectrum raw, spectrum processed and individual peak fits should all be instances of the same Spectrum Class.

Baselines can be inferred by subtracting raw minus processed spectra. Total fits can the inferred by summing all peak fits.