RealizationSelection plugin modification

[gavinevans]

Related to https://github.com/metoppv/mo-blue-team/issues/783

Description
This PR makes a number of updates to the RealizationSelection plugin following trying to use the plugin with more realistic data:

• Building on changes made in Support resetting the forecast_reference_time when clustering #2359, the forecast_reference_time and forecast_period on the input is reset to be from a common forecast_reference_time.
• Add an exception is no forecast cubes are provided as an input.
• Alter how the realizations are selected from the model cubes, so that the index of the realization is used, rather than the realization number on the coordinate.
• Add the "cluster_sources" attribute to the result cube, from the cluster_cube, as this attribute is expected by the ForecastTrajectoryGapFiller plugin.
• Alter how the name of the primary model is identified. Now the cluster_sources attribute is used, rather than expecting a "mosg__model_configuration" to be present on the cluster_cube. This attribute wouldn't be expected to be present on the cluster_cube.
• Alter how the nearest secondary input is identified. This was an issue, for example, if one secondary input finished at T+12, the T+12:15 forecast would get mapped to using the secondary input from T+12, even though that input is not available at T+12:15. Therefore, the nearest, larger forecast period is now identified e.g. T+13, and the mapping from this forecast period is used.

Testing:

• Ran tests and they passed OK
• Added new tests for the new feature(s)