cherry-pick: Honor branchedDatabases for package-deployed applications (conflicts → v5.3)

[github-actions]

Cherry-pick of PR #3083 onto `v5.3` produced conflicts on commit(s): `834835f5351b421fcbf1dae08f8da2e64a8a13c0 4f1aed8`.

Resolve the conflict markers on branch `cherry-pick/v5.3/pr-3083` and merge this PR.

@claude please review branch `cherry-pick/v5.3/pr-3083` and suggest a patch that resolves the conflict markers (<<<<<<< / ======= / >>>>>>>) introduced by cherry-picking PR #3083 onto `v5.3`. Post the suggested patch as a comment here — do not push.

Resolution on v5.3

The conflict markers are resolved in a new commit on top of the Action's commits. No history was rewritten.

• DESIGN.md index. Kept v5.3's own entries for the deploy and the serialized preparation. The canary wording in them is main-only. The package-branch entry appears once, at the end of the components section, where Honor branchedDatabases for package-deployed applications #3083's second commit placed it.
• componentLoader.ts. Both options are kept on a root package's own load: branchedDatabases from #3083 and v5.3's collectScopes. The first is set only when isRoot is true. The second is set only by v5.3's non-root deploy validation load, so the two never meet. Every production hunk now matches the squash of Honor branchedDatabases for package-deployed applications #3083 on main (e72eaaf).

Adapted to v5.3 (the cherry-picked tests assumed main's canary)

v5.3 has no deploy canary. An operations-API deploy answers 200 without a load check, and a refused package fails only when the restarted worker loads it.

• The integration test's refused-deploy helper asserts the 200 and then waits for the refusal itself. For the missing database, that is the route going 404. For the root plugin package, it is get_status reporting error. The no-base-write and no-fork assertions are unchanged. The certification assertions and the isolated redeploy (which existed only to reach a certifiable load) were dropped.
• The unit test's misplaced-config case asserts the ErrorResource. v5.3's outer loadComponent catch records no component status.
• The components/DESIGN.md note drops the canary references and states how v5.3 reports these failures.

❓ Your call: on v5.3, a package whose branch preparation fails (for example a missing database) serves no routes and logs the error, but get_status reads it loaded. The nested load's catch records no status, and the root loop then marks an application with no plugin module as loaded. Main fixed both halves inside the canary PR (c3eb2f2). Other swallowed package-load failures on v5.3, such as a config-parse error or a misplaced branchedDatabases, already report the same way. I left it out of this backport and documented it, because fixing it changes get_status for every failed application load on the 5.3 line. If you want it, it is two small hunks backported from main as a separate v5.3 PR.

Verification

• CI was dispatched on 2ecb1b8ef: Integration Tests (Node 24) ✅ and Unit Test (Node 24) ✅. The new test passed on the Node 24, uWS and Bun legs and is skipped on Windows by design.
• Locally (RocksDB): componentLoaderBranches 4/4 and branched-database-package 8/8. Against v5.3's original componentLoader.ts, the adapted integration test fails 8/8, each at the intended assertion, so the v5.3 assertions are not vacuous.
• The PR-triggered Unit Test (Node.js v26) fails with withNodeAdapter compression and install_node_modules dry-run failures on Node 26.11.1. The same failures appear on other open PRs on both main and v5.3, so they are not from this change.
• Pre-push review (codex, gemini, cursor-composer, harper-domain; 3 rounds): 2 findings fixed. 1 minor is deferred as the call above. The remaining nits concern text that is byte-identical to main's Honor branchedDatabases for package-deployed applications #3083.

— Claude Opus 5.5

🤖 Generated with Claude Code

https://claude.ai/code/session_011KV6qxwDSB78CKhegmi8Rf

Related PRs: #3083 overlaps, 5 others independent

_{Review-Coverage: authored=claude; ran=cursor-composer,gemini,codex; adjudicated=domain; declined=cursor-grok,cursor-kimi,cursor-muse; rounds=3; full=1 @ 2ecb1b8}

_{Review-Attention: read ~3m (decisions: document-not-fix-status-gap, refuse-branched-root-plugin, refusal-after-preparation) @ 2ecb1b8}