Better sampling of sigma_8

[tilmantroester]

This PR replaces the janky sigma_8 sampling setup. Instead of running CAMB twice, the transfer function (which takes the most time) is only computed once and then the power spectra are recomputed using the correct A_s corresponding to the requested sigma_8.

This speeds things up by ~50% and also avoids the numerical uncertainty in the initial estimate of sigma_8 due to the low accuracy settings of the first run. Since sigma_8_input now agrees with the output sigma_8, I added the option of taking sigma_8 as the input directly. If both sigma_8 and A_s are provided, sigma_8 takes precedence.

This approach could in principle be used to do fast sampling of all non-cosmology parameters, such as primordial power spectrum parameters in general or HMCode parameters. See for example cells 50 onward here: https://camb.readthedocs.io/en/latest/CAMBdemo.html

[joezuntz]

One more request - could you update e.g the des-y1 ini file to use the new name. That is tested in the CI.

[tilmantroester]

Back to you!

[joezuntz]

Brilliant, thanks for this!