Implement frequency-domain bootstrapping

[astrogilda]

Is your feature request related to a problem? Please describe.
All extant methods are in the original time series domain. The package could benefit by adding complementary frequency-domain methods.

Describe the solution you'd like
Some helpful papers (evolving list):
1
2

[fkiraly]

To me, this (at least the non-paywalled paper 2) reads like a pipeline or composite. That is transform -> bootstrap -> inverse_transform, where transform is fft, or transform (fft) -> transform (smoothe) -> bootstrap -> inverse_transform(fft)

Should we read this as a case for pipelines?

At least in sktime, where fft is already present, this pipeline could be easily built.