Allow arbitrary IAM function in pvlib.iam.marion_diffuse

[markcampanelli]

Is your feature request related to a problem? Please describe.
I am working on an application using pvlib.iam.marion_diffuse, and I would like to inject my own IAM function (from interpolated IEC 61853-2 data) as a parameter to pvlib.iam.marion_diffuse.

Describe the solution you'd like
Add the option for passing a callable to pvlib.iam.marion_diffuse instead of the existing strings for the builtin IAM model functions. (The existing strings would still be accepted.)

Also add a function to create IAM functions as a PCHIP interpolation of data such as from IEC 61853-2. Note that this solution can also be made to add support for passing the existing pvlib.iam.interp function (whose cubic option is slightly different that pchip) to pvlib.iam.marion_diffuse.

Describe alternatives you've considered
Rolling my own interpolant each time, and then calling the lower-level pvlib.iam.marion_integrate for each of the three components.

Additional context
This is a followup to #1980, but without any concern to improving the diffuse-horizontal integral.

[adriesse]

Have you considered adding the pchip option into pvlib.iam.interp rather than creating a new function?

[markcampanelli]

@adriesse

Have you considered adding the pchip option into pvlib.iam.interp rather than creating a new function?

I am open to this consolidation. Do you think folks would be open to pvlib.iam.interp returning a callable that could then be passed to the new pvlib.iam.marion_diffuse (or used for other calculations)?

If I understand correctly, this avoids some overhead of recreating the interpolating function on each usage, although it would break the existing interface to pvlib.iam.interp.

[adriesse]

@adriesse

Have you considered adding the pchip option into pvlib.iam.interp rather than creating a new function?

I am open to this consolidation. Do you think folks would be open to pvlib.iam.interp returning a callable that could then be passed to the new pvlib.iam.marion_diffuse (or used for other calculations)?

If I understand correctly, this avoids some overhead of recreating the interpolating function on each usage, although it would break the existing interface to pvlib.iam.interp.

I think Marion only has one call to the model.

[markcampanelli]

I noticed that the builtin interp and schlick AOI modifiers didn't have full support in ModelChain like the other builtins, so I'm trying to add that here too.

[echedey-ls]

Feel free to break down these issues into different PRs, doing too many things in one big PR makes the review process more time consuming - and reviewers more hesitant to review.