Flox

- `engine="flox"` is now the default
- multiple bugfixes.

- Support method="cohorts", "blockwise", "split-reduce" for nD `by`
- Add engine="flox" which brings substantial performance improvements (but still a little buggy)
- Deleted xarray_groupby_reduce
- Multiple bugfixes

Minimized dependencies
Some performance improvements.

Many bugfixes and renamed to flox