Fix covariate adjustment looking up preenrollment table by metric name instead of data source

[scholtzan]

In discrete-metrics mode, metric tables are partitioned by data source ({slug}_{basis}_{data_source_name}_{period}_{idx}), with each metric as a column. But covariate adjustment looked up the preenrollment table by metric name, so the table never existed and adjustment silently fell back to unadjusted inferences every time.

Fixed both the query builder (_create_subset_metric_table_query_covariate) and the Dask prerequisite computation (_subset_metric_table_prerequisites) to resolve the table by data source name. Since a covariate can be a different metric than the during-experiment metric, the new _covariate_table_metric_name helper looks up the covariate metric's own data source.