WIP: feat time domain solar wind covariance

[jeremy-baier]

Needed for NG20.
Major changes

• Making phi either 1d or 2d. Necessary for 2-dimensional kernels.
• Addition of said 2d kernels for the solar wind only for now.
• Time domain interpolation bases to go with the above. Both custom ( for Bayesian Blocking ) and uniform bins.
• Since above, need to switch trivia phi inversion to use a cholesky decomposition with a np inversion fallback.

More work should really explore additional caching of various parts of the GLS fit when the noise model is being held constant. This might be sub blocks of the design matrix. Or a refactorization of the GLS implementation.

TODO

• unit tests
• should check that our basis and phi matrix match those of discovery. (might consider adding tests for this).
• additional model validations ?
• Try refactoring so there is one TimeDomainSolarWindNoise class with a kernel parameter

Disclosure that I used co-pilot for generating the unit tests as well as some docstrings and type hinting.

[scottransom]

There's a lot in here and it is beyond me to go through carefully line-by-line! We definitely need to make sure that there are some test cases and/or examples available that provide results that you believe are correct.

[jeremy-baier]

thanks for the comments, Scott ! I will take a look and see about implementing them.

[dlakaplan]

This could use a slightly more explicit docstring (explaining parameters and return values)

[dlakaplan]

(also I see some typos in utils.py, so you could fix paremeters -> parameters

[dlakaplan]

Maybe @abhisrkckl can look at this, since it involves changes to the noise model and some of the whitened residual pieces he put in recently?

[abhisrkckl]

Is this ready to be looked at?

[jeremy-baier]

Is this ready to be looked at?

Yes ! I just need to get the black to pass. I think I have been using the wrong line length settings.