ensure there are no dependency loops in our workflows

[tomjemmett]

Currently, we generate the reference schema, then the raw_data schema, then the aggregated_data (and, by extension, default) schema, and finally the inputs_data schema.

In create_population_by_imd_decile() though, which is part of the reference schema, we use the default schema's data.

While in practice this does work (as the tables now exist), ideally we should not have any functional dependencies like this.

We might need to split the reference workflow into pre raw_data and post raw_data steps