[CT-2691] Populate MetricNode depends_on

[QMalcolm]

No description provided.

[jtcohen6]

@QMalcolm Am I right in thinking that this would require dbt to traverse metric → measure → semantic_model, so that we can record SemMod unique IDs in the metric's depends_on.nodes?

A metric's filter can also depend on a dimension from a semantic model that is different from the semantic model it depends on via its measure.

I think this would be a prerequisite for DAG relationships to work:

• dbt list -s my_semantic_model+ to include metrics
• DAG visualization in dbt-docs ([CT-2740] Ensure semantic models & metrics are included in DAG viz dbt-docs#431)

[QMalcolm]

I chatted with @jtcohen6 earlier. There are three-ish places that metrics need to source the nodes for it's depends_on attribute

1. Metric.type_params.input_metrics for RATIO and DERIVED type metrics
2. Metric.type_params.measure for SIMPLE and CUMULATIVE type metrics
3. From filters
   a. Metric.filter
   b. Metric.type_params.measure.filter
   c. Metric.type_params.input_metrics[x].filter

The work for 1 and 2 exist currently on #8015.

Number 3 is a bit harder though. This is because
A. The spec for filters actually isn't finalized yet. The MetricFlow team is working on support for dunders
B. Which semantic model a dimension comes from in a filter is determined at query time (because a dimension can exist on multiple semantic models and the optimal join path is calculated).

Thus this issue is just to resolve 1 & 2, whereas 3 will be pushed off and become a separate core ticket/issue