Extrapolating model atmosphere quantities off their definition range

[aprsa]

When computing model atmosphere quantities, namely normal emergent passband intensity (Inorm), specific emergent passband intensity (Imu), limb darkening coefficients (ld_coeffs), integrated limb darkening model (ldint), extinction factors and boosting factors, we need to prescribe how to extrapolate. There are currently three supported modes: none, nearest and linear, which should be self-explanatory. In addition, we make a distinction between intensities and limb darkening -- there is no fundamental reason why both should be extrapolated the same way. In fact, it is typically appropriate to extrapolate intensities linearly and approximate LD coefficients/integrals with the nearest on-grid value.

The current frontend implementation defines two parameters, blending_method and ld_blending_method. The "blending" part is really a misnomer, stemming from the times where these two parameters were considered just in the context of blending. The more appropriate naming convention would arguably be atm_extrapolation_method and ld_extrapolation_method. This is (partly) reflected in the passbands backend. It might be tempting to consider intens_extrapolation_method in place of atm_extrapolation_method, but that does not work well for boosting, extinction, etc. Instead, atm catches all those quantities.

The more pertinent problem is that, in the current implementation, ld_blending_method is exposed only if blending_method != 'none'. While this in principle sounds reasonable -- if we are not extrapolating intensities, we probably should not be extrapolating LD quantities either -- there is no fundamental reason to prevent it; this too stems back to the times when blending was the considered to be the only circumstance when anyone would want to extrapolate off-grid. However, when calling only LD functions, blending_method is not even used, and ld_blending_method cannot be assigned/changed from 'none' because of its dependency on blending_method != 'none'. This is arguably overly constrictive.

My suggestion is to remove the dependency of ld_blending_method on blending_method and to rename both parameters into atm_extrapolation_method and ld_extrapolation_method, respectively. Any implications on blending should be considered secondary to this change and should be dealt with on the blending side.