Improve extrapolation in Pk2D

[jablazek]

Extend the functionality built-in to Pk2D to allow for power-law (or similar) extrapolation when is_logp = False. This will prevent differences in Pk that comes from nl_pt vs directly from the main cosmology object, and it will provide a physically reasonable extrapolation for PT calculations. This closes #773, since it will solve the underlying issue.

Something like this:

if is_logp = True continue as normal (could add an option for all neg values, i.e. log(-P) was passed)

if is_logp = False if log_ext = False(new option) ... original behavior if log_ext = True

• use the last two values on both ends (or three, depending on desired order of taylor expansion. or use more if we are worried about noise)
• confirm all positive or negative
• take log of P (or -P)
• perform 1st or 2nd order extrapolation
• (could also use the original Pk2D extension for all pos or all neg values, but we might want to allow a purely flexible power law extension without assuming n_s at low k)

[jablazek]

@tilmantroester : will this also solve the nonlinear Pk extrapolation for nonlinear systematics in Firecrown?