Fix the incremental mask computation in the steps blending method

[ladc]

In the extrapolation based nowcasting, the mask (NB: not the radar mask but rather the rain-norain mask in which we allow the generation of new rainfall by the noise process) is dilated iteratively at every timestep. This is easy to do in Lagrangian space for a pure extrapolation-based nowcast. However, in the case of STEPS blending with NWP, this is less straightforward because we are no longer in Lagrangian space (NWP is not advected). Right now it's dilated only once but this does not grow further per timestep. This should be adapted but for that the rain mask probably needs to be recalculated and dilated every timestep.