Support Multi-Scope Ratio Metrics

[shangyian]

Problem

DJ doesn't have a clean way to define metrics that operate at different granularities than the query grain. Common use cases that are difficult today:

1. Share of total - numerator respects filters, denominator is global
2. Market share within category - compute parent-level totals while querying at child level

Proposed Solution

Add two new fields to metrics: fixed and exclude.

  metrics:
    - name: revenue
      expression: SUM(revenue)

    - name: total_revenue
      expression: SUM(revenue)
      fixed: true                    # Ignores all dimensions/filters

    - name: genre_revenue
      expression: SUM(revenue)
      exclude: [title]               # Ignores title + hierarchy children

    - name: daily_active_users
      expression: COUNT(DISTINCT user_id)
      required_dimensions: [date]    # Existing - must include date in query

Field Def	Type	Behavior
required_dimensions	list[str]	Must include these dimensions
exclude	list[str]	Ignore these dimensions + hierarchy children
fixed	bool	Ignore all dimensions and filters (global)

Example: Market Share Within Genre

  metrics:
    - name: revenue
      expression: SUM(revenue)

    - name: genre_revenue
      expression: SUM(revenue)
      exclude: [title]

    - name: market_share_within_genre
      expression: revenue / genre_revenue
      type: derived

Query at title + genre grain yields generated SQL:

  SELECT
      title,
      genre,
      SUM(revenue) as revenue,
      SUM(revenue) OVER (PARTITION BY genre) as genre_revenue,
      SUM(revenue) / SUM(revenue) OVER (PARTITION BY genre) as market_share_within_genre
  FROM fact_table
  GROUP BY title, genre

Hierarchy Behavior for exclude

When excluding a dimension, its children in the hierarchy are also excluded:

Hierarchy: genre -> title -> episode

exclude: [title] means:
- title: EXCLUDED
- episode: EXCLUDED (child of title)
- genre: KEPT (parent of title)