fix(ci): e2e gates always emit a result so auto-promote can read it

[HongmingWang-Rabbit]

Summary

auto-promote-staging.yml's gate-check at line 99 treats "workflow didn't run" the same as "workflow failed":

if [ "$RESULT" != "completed/success" ]; then
  ALL_GREEN=false
fi

E2E Staging Canvas (Playwright) and E2E API Smoke Test both have paths: filters on their triggers. A push that doesn't touch the watched paths never triggers the workflow, so auto-promote sees missing/none, marks all_green=false, and aborts. Every platform-only or test-only push dead-locks here. This bit us today on PR #2201 (touched only tests/e2e/test_staging_full_saas.sh) — required a manual staging → main bridge (PR #2202) to unstick.

Design B — always-run + fast-skip

For each gate workflow:

• Drop paths: from the push: and pull_request: triggers
• Add a detect-changes job using dorny/paths-filter@v3, scoped to the same paths the trigger filter used to watch
• Gate the real-work job on needs.detect-changes.outputs.<filter> matching true
• Add a sibling no-op job that runs when the filter output is false and exits success, so the workflow run's conclusion is success either way

workflow_dispatch and the weekly canvas schedule short-circuit detect-changes to always-run — those triggers exist to exercise the suite, not to be silently no-op'd.

Why this over Design A (smarter auto-promote)

Design A (considered + rejected): teach auto-promote-staging.yml to read each gate's paths: filter and treat "no run because filter excluded the commit" as conditional pass. Couples auto-promote to other workflows' YAML schema; breaks silently if a gate is renamed or its filter changes. Design B keeps the auto-promote contract simple ("each gate emits success") and makes each gate self-describing.

Aligns with the user-stated principle: long-term, robust, fully automated, eliminate human error.

Cost

~10-30s of runner time per gate per push for the no-op job when paths don't match. Negligible vs dead-locked promotion chains.

Test plan

• yaml syntax validates
• After merge: a non-canvas, non-workspace-server push to staging should produce green runs for both gates (no-op pass), and auto-promote should fast-forward main
• A canvas-touching push: e2e-staging-canvas runs the playwright job (real work)
• A workspace-server-touching push: e2e-api runs the e2e-api job (real work)

Follow-ups (separate PRs)

• e2e-staging-saas.yml has the same path-filter pattern but isn't currently in the auto-promote gate list — patch it preventively for consistency, or wait until it becomes a gate
• auto-promote-on-e2e.yml for :staging-<sha> → :latest retag (Design B unblocks this — the auto-promote chain becomes deterministic)

🤖 Generated with Claude Code