Speed up GraphQL findNodes via batched extract

[shangyian]

Summary

This PR cuts findNodes(nodeTypes: [METRIC]) wall-clock from ~14s to ~1s on a 1000-metric fragment that reads extractedMeasures, metricMetadata, and columns. The resolver now serves the common case from denormalized columns instead of parsing and extracting per-metric on every request.

• extractedMeasures fast path + batching:
  • When the fragment on extractedMeasures only reads derivedQuery, return NodeRevision.derived_expression (which gets populated by the derive_frozen_measures background task on node create/update) and skip MetricComponentExtractor.extract(). This reduces the number of ANTLR parses and AST walks.
  • It falls through to the full path when components, combiner, or the derivedExpression alias is requested. When the full path is needed, a request-scoped loader batches N per-metric extractions into one session.
• Drops redundant joinedload(Column.dimension/partition), which are already lazy="joined" on the relationship and were causing cartesian row fanout during hydration.
• Conditional metricMetadata AST parse. Skip the ANTLR parse + function walk unless the fragment requests expression or incompatibleDruidFunctions.

Also added a backfill scrip that populates derived_expression for metrics created before derive_frozen_measures existed, so they hit the fast path after deploy.

Test Plan

• PR has an associated issue: #
• make check passes
• make test shows 100% unit test coverage

Deployment Plan

[netlify]

✅ Deploy Preview for thriving-cassata-78ae72 canceled.

Name	Link
🔨 Latest commit	ea79599
🔍 Latest deploy log	https://app.netlify.com/projects/thriving-cassata-78ae72/deploys/69eb2e81c28be70008141973