Feature extraction for different intervals in `PipeOpFFS` about mlr3fda HOT 3 CLOSED

The question here would which form custom functions should have that are passed to the PipeOp.
Maybe instead of having them be function(arg, value) we instead want them to be function(arg, value, lower, upper) where arg and value are the full vectors and lower and upper the lower and upper bound(s).
Then the custom function is responsible for how to calculate the feature for all the specified intervals.
We could make the feature extraction for different intervals then very efficient.

from mlr3fda.

In addition to specifying multiple lower and upper bounds, one might also want to specify a lower and upper per extracted feature (e.g. differing windows for mean and variance).
While one can in principle also do this with different PipeOps, there is a speed disadvantage.

from mlr3fda.

Suggestion:

We have a parameter interval instead of lower and upper that can either be a 2d vector or a named list, e.g.list(mean = c(0, 10), var = c(0, 20)).

If we then also want to allow multiple windows per feature we can have something like

interval = list(mean = list(c(0, 1), c(1, 2)), var = list(c(10, 20), c(20, 30)))

from mlr3fda.