Rethinking chunking module

[jonasvdd]

The chunking submodule is an amazing feature of tsflex, however it still has some flaws.

It assumes:

• [ ] that each series-signal is sampled at a (nearly) fixed frequency) -> hence the fs_dict parameter
• [ ] has way to many complexity within a single method (hard to read the code and improve)
• [x] does not conveniently support the flexibility of only performing time-based chunking

With this issue, I hope to openly discuss how others see this challenge

[jonasvdd]

Some additional notes:

• Isn't chunking a variant of a constrained segmenter
• As we don't support multi-indexed data, I think group-based chunking support should be in the roadmap

[jonasvdd]

See also this gist: https://gist.github.com/mcourteaux/fc6edd7feaa180da224782933a7b6079