chore(ai-reviews): bump ai-workflows pin to v3.1.4 (silent-failure fix + clear outcome signaling)

Summary

Bump the dotCMS/ai-workflows pin in the AI review workflows from @v3.1.2 to @v3.1.3.

Why

~43% of recent GPT-5.5 automatic reviews were posting "❌ Codex Review failed — job failed before producing output" and giving the PR no review. Root cause (fixed in ai-workflows#38, released as v3.1.3): on the Responses API, max_output_tokens caps the combined reasoning + answer tokens, so at reasoning_effort: medium GPT-5.5 sometimes spent the whole 2048 budget reasoning and returned status=incomplete with no visible text — which the executor treated as a hard failure.

v3.1.3 raises the default budget to 8000, captures the incomplete terminal event, retries once with a larger budget + lighter reasoning, and posts a clear diagnostic instead of a generic failure.

Validated e2e (before/after) on dotCMS/steve-quarterly-planning #105.

Changes

• ai_claude-orchestrator.yml: @v3.1.2 → @v3.1.3 (both jobs)
• ai_claude-backend-reviewer.yml: @v3.1.2 → @v3.1.3

Start with ai_claude-orchestrator.yml and ai_claude-backend-reviewer.yml, locating the referenced dotCMS/ai-workflows pins in both orchestrator jobs and the backend reviewer. Resolve the version discrepancy between the title's v3.1.4 and the body and changes' v3.1.3, then verify both workflow files use the intended release.