workflow run list times out on some organizations

[migmartri]

Problem

Listing workflow runs (/workflow-runs) times out for some organizations. The slow query is the org-scoped workflow_runs list:

SELECT workflow_runs.* FROM workflow_runs
WHERE EXISTS (
  SELECT workflows.id FROM workflows
  WHERE workflow_runs.workflow_id = workflows.id
    AND workflows.deleted_at IS NULL
    AND workflows.organization_id = $1
)
ORDER BY workflow_runs.created_at DESC
LIMIT 26;

Observed runtime: ~10s (request times out) on affected organizations. Other orgs are fast.

Root cause

workflow_runs has no organization_id column, so org scoping has to traverse the workflows edge. With the correlated EXISTS form combined with ORDER BY created_at DESC LIMIT 26, the planner has two main options:

1. Backward index scan on workflow_runs.created_at, checking the EXISTS subquery for every row.
2. Nested loop through workflow IDs of the org, then merge by created_at.

For organizations with sparse recent activity but many workflow_runs rows from other orgs in front of theirs by created_at, the planner picks option 1 and walks most of the table before finding 26 matches. Hence the timeout.

A prior fix (#1635) addressed a related shape by reordering composite indexes. That fix only helps when the planner picks the workflow-driven plan; the current EXISTS shape lets it regress to the backward index scan.

Fix direction

Denormalize organization_id onto workflow_runs and add an (organization_id, created_at DESC) index so the org-scoped list becomes a single sargable range scan, independent of how many workflows the org has.