Add Kalman-filter based post-processing to blending scheme

[RubenImhoff]

As discussed at the start of our blending implementation and re-discussed during this year's EGU, we can still improve the post-processing that takes place in the blending procedure. Right now, the post-processing procedures from the nowcasting module are used and adjusted to work within the blending module. This is functional, but has as limitation that the post-processing is based on (1) having everything in Lagrangian coordinates and that is not the case in the blending procedure (we have incorporated that difference, but is suboptimal), and (2) it depends on the weights assigned to the NWP and extrapolation component, which could reduce extremes.

A Kalman-filter like method, such as in Nerini et al. (2019; https://doi.org/10.1175/MWR-D-18-0258.1), could do the trick. I think we should see if we can implement this in the blending module anytime soon. :)

[dnerini]

Thanks for initiating this @RubenImhoff ! Can you provide any hint on where we should make changes in the current code? I can then try to make a first draft.