Add IsotropicNoiseEstimator

[janden]

This would essentially wrap (or inherit from?) the AnisotropicNoiseEstimator and simply radially average the result to obtain a circularly symmetric power spectrum estimate. This radial averaging can be done, for example, by projecting onto an FB basis, zeroing out all coefficients corresponding to non-zero angular frequencies, and evaluating the resulting expansion. This is similar to what's done in the Cov2D.get_mean method.

[junchaoxia]

There is a new issue #323 opened for the IsotropicNoiseEstimator. I think this one is redundant. Need to close it when #323 is done.

[garrettwrong]

Working on this now. Have ported epsdS and am trying to integrate it with the NoiseEstimator now. There were existing tests in the MATLAB code I will look into porting as well.

We can consider alternative implementations after porting the legacy one that should support the JSB experiment repros.

[garrettwrong]

Still working on that integration. Here is an example PSD from the epsdS port.

[garrettwrong]

#1223 implements the MATLAB version.

[garrettwrong]

I'm moving this task to done in 0.13.2 because that release adds an isotropic whitening method via legacy_whiten.

However, I am leaving the issue open as there is still the task of creating a more optimal isotropic whitening method by extending out existing aniso tooling. This was the original task, and should be done in a future release.