extrapolation and other numerical / unary operations

[mikapfl]

compose generally does binary operations, i.e. a given timeseries is combined with
a second timeseries to yield a better result.

We also need support for unary operations, where a given timeseries is handled
(e.g. by interpolation or extrapolation) to yield a better result.

Possible unary operations:

• extrapolation
• interpolation
• extrapolation with proxy data
• smoothing
• de-trending
• replacing by NaNs

Some of these functions already exist in xarray or primap2 (e.g. replacing by NaNs
can be done using pr.set), others are special for our use case. The most urgent ones
are extrapolation and interpolation.

We expect that we will use these functions interactively / ad-hoc, so it would be
useful to have them as pr. functions first. In a second step, we likely also want
to have more complex configs (e.g. for full primap-hist processing) which might be
something we want to do in a similar fashion as the compose function.