perf: Denormalise feature names on Code Reference scans and add GIN index

[matthewelwell]

Changes

Closes #6836

This PR denormalises the feature names directly into an ArrayField on the FeatureFlagCodeReferencesScan. The feature names were previously nested inside the JSONField which caused performance issues with the database queries when listing features for the dashboard.

How did you test this code?

Added tests.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

3 Skipped Deployments

Project	Deployment	Actions	Updated (UTC)
docs	Ignored	Preview	Mar 4, 2026 6:09pm
flagsmith-frontend-preview	Ignored	Preview	Mar 4, 2026 6:09pm
flagsmith-frontend-staging	Ignored	Preview	Mar 4, 2026 6:09pm

[github-actions]

Docker builds report

Image	Build Status	Security report
ghcr.io/flagsmith/flagsmith-e2e:pr-6844	Finished ✅	Skipped
ghcr.io/flagsmith/flagsmith-api-test:pr-6844	Finished ✅	Skipped
ghcr.io/flagsmith/flagsmith-frontend:pr-6844	Finished ✅	Results ✅
ghcr.io/flagsmith/flagsmith-api:pr-6844	Finished ✅	Results ✅
ghcr.io/flagsmith/flagsmith:pr-6844	Finished ✅	Results ✅
ghcr.io/flagsmith/flagsmith-private-cloud:pr-6844	Finished ✅	Results ✅

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.34%. Comparing base (800ac8d) to head (2b03644).
⚠️ Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6844      +/-   ##
==========================================
+ Coverage   98.32%   98.34%   +0.01%     
==========================================
  Files        1332     1342      +10     
  Lines       49401    49723     +322     
==========================================
+ Hits        48575    48898     +323     
+ Misses        826      825       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[matthewelwell]

I'm fairly sure there are other fields in this app that don't support oracle, but we probably need to think on how to exclude this from EE.

[gagantrivedi]

The PR migrates the last_feature_found_at subquery and the get_code_references_for_feature_flag function to use the new feature_names array, but the count calculation in
  counts_by_repository (the heavier operation) still does a full jsonb_path_query_array + jsonb_array_length against the raw code_references JSON blob:

  count=Func(
      Func(
          F("code_references"),
          Value("$[*] ? (@.feature_name == $feature_name)"),
          JSONObject(feature_name=F("feature_name")),
          function="jsonb_path_query_array",
      ),
      function="jsonb_array_length",
  ),

  This generates SQL like:

  jsonb_array_length(
    jsonb_path_query_array(
      V0."code_references",
      '$[*] ? (@.feature_name == $feature_name)',
      JSONB_BUILD_OBJECT('feature_name', "features_feature"."name")
    )
  )

The only solution I see here is to accept that the data model doesn't work for how we query it, and change it to flatten everything and get rid of the JSON column. Otherwise, we'd keep going in circles making the data access layer more complicated than it needs to be.