Allow more flexible CosmoPower network settings.

[HTJense]

CosmoPower as it currently stands requires the user to include a TT/TE/EE network of NN/PCAplusNN/NN type. Since this is a pretty strict set of requirements, this commit works on changing that.

Instead of each network being defined only by a network path, there is now a network_settings dict within the CosmoPower theory block, which lists each network and what type of network it specifies. This allows one to:

• include more or fewer networks than just TT/TE/EE (think inclusion of BB or pp).
• Added support for pp.
• specify the type of network for each spectrum. No longer is TE specified to be a PCA+NN type, but one can specify the network type for each network individually.
• The required input parameters are now checked at initialization instead.
• (4400ef0) If your network has been trained on D_ell instead of C_ell, you can set a flag that will take care of the prefactor (a necessity, based on personal experience).
• (cdf3e80) Added networks for derived parameter calculation.

Some things that might be worthwhile adding before or after this merge include:

• Ensure that networks trained over different ell-ranges are supported (right now it still asserts every network operates over the same ell range, but this may not be necessarily true, see also point below).
• P(k), see also Add CosmoPower P(k, z) emulation  #84.
• Derived parameter networks (see Boris Bolliet's work). Added in cdf3e80.
• A better pytest implementation.

[HTJense]

Current state is good enough to go into main branch for me, it already allows for an improved experience by end-users.
Extra features (i.e. non-Cl spectra, derived parameters, better error coverage, etc.) is probably more fitting for a different pull request, to prevent lengthy review & development hell.

[itrharrison]

Looks good, just need to mention that the ell_factors are for CLASS units.

[HTJense]

Any updates on this @itrharrison ?

Done.

[itrharrison]

It would be good to have a more extensive test suite, but I am happy with this for now.