Add spatial morphing inputs and kgos

[gavinevans]

Related to metoppv/improver#2441

This PR adds inputs and kgos associated with the SpatialMorphing plugin and CLI. The plot below shows the input and outputs from the SpatialMorphing CLI. source a is a nowcast, source b is a MOGREPS-UK realization. The "no_implied_transitions_kgo.nc" shows the result where the json file defining the transitions has been modified, so that the forecast period between processed here is after the transition period, meaning that the MOGREPS-UK realization is returned. The known good outputs show the output at an interpolated timestep (with and without quantile mapping) when a linear morphing method is used. This linear morphing method is included for demonstration as Google FILM is a bit heavier, computationally to run as part of the acceptance tests. Due to the use of linear weighting and how the quantile mapping is configured (also using a linear blend of the forecasts as a reference), the quantile mapping has not effect.

The plot below shows the same results when Google FILM is used as the morphing method, rather than using a linear approach. The Google FILM morphing method generally results in reduced extent to the precipitation, which would be the hoped for result, given that the linear blending approach will retain the spatial extent of the precipitation found in the two forecast sources. In this case, the quantile mapping has a clear impact with some evidence that it has further slightly reduced the abundance of precipitating pixels.

For completeness, I've added testing that smoothstep transition weights scheme, which results in no discernible difference relative to the linear transition weights scheme in this example.