fix(ci): auto-review bot-opened PRs, drop manual review dispatch

[absorbb]

Problem

AI Review fails on the automated security-fix PRs (e.g. security/fix-npm-2026-06-22). The 🤖 Run Codex analysis step aborts with:

Error: Actor 'github-actions[bot]' is not permitted to run this action:
Actor 'github-actions[bot]' must have write access to jitsucom/jitsu.
Detected permission: 'none'.

openai/codex-action's write-access pre-check queries the collaborators API, which returns none for the github-actions[bot] actor, so the action exits 1. The previous skip-guard in ai-review.yml checked for jitsu-code-review[bot] — but these PRs are opened under the repo's own GITHUB_TOKEN, i.e. actor github-actions[bot], so the guard never matched and the auto-trigger ran and failed.

Fix

The reusable workflow now passes allow-bots: true to codex-action (jitsucom/github-workflows@1.29.20260622), which whitelists the GitHub-owned github-actions[bot] specifically. The review still posts via the AI_CODE_REVIEW app token (PR write) — the actor check was the only gate.

This PR adopts that on the jitsu side:

• ai-review.yml: bump the reusable workflow ref 1.28.20260529 → 1.29.20260622, and drop the dead skip-guard so bot-opened PRs are reviewed. Draft PRs are still skipped inside the reusable workflow.
• security-fix.yml: remove the trigger-review job. It manually dispatched the review as a workaround for the failing auto-trigger; now that the auto-trigger works, it would only produce a duplicate review per security PR.

Upstream change

jitsucom/github-workflows@1.29.20260622 — adds allow-bots: true to the codex-action step. allow-bots only trusts github-actions[bot] (repo-internal automation running under the repo's own token), not arbitrary external actors.

🤖 Generated with Claude Code

[jitsu-code-review]

Reviewed .github/workflows/ai-review.yml and .github/workflows/security-fix.yml for workflow correctness and security impact.

The changes are coherent: the reusable AI-review workflow version now supports bot actors, and removing the manual dispatch from security-fix.yml aligns with the existing PR-triggered AI review path to avoid duplicate review runs. I did not find actionable correctness or security regressions in this patch.