More efficient multi-dimensional grids

[stephenmwilkins]

This is a version 2 challenge but flagging now since it will need considerable development.

In the future we will have grids that potentially have much higher dimensionality, with multiple photoionisation parameters. This posses a challenge because the grids could become very large, there will also be significant repetition. For example, the same incident spectra may exist multiple times.

I think this can be addressed with the following steps:

• store the incident spectra with only the incident aces (e.g. age and metallicity). When extracting spectra only the incident axes would be used.
• compress the nebular_continuum spectra using PCA, NNMF etc. In the context of PCA (probably not the best approach) the grid would contain the principle components and then the parameters at every (incident + photoionisation) grid point.
• construct the linecont spectra from a LineCollection.
• construct the nebular spectra from nebular_continuum and linecont.
• calculate the transmission =transmitted/incident and compress. The rationale here is that the transmission should be relatively simple.
• construct the transmitted spectra from the transmission and incident.