chore(ai-reviews): bump ai-workflows to v3.1.5 + switch auto-review to DeepSeek V3.2 (gpt-5.5 down on mantle)

[sfreudenthaler]

Bump dotCMS/ai-workflows pin v3.1.4 → v3.1.5 in the AI review workflows.

v3.1.5 (ai-workflows#40) surfaces the actual failure reason when the mantle Responses API returns response.failed (it was discarded before — the sticky just said "failed before producing output"). Now the ❌ sticky + job log show response.error (e.g. server_error), notes it's typically an AWS/mantle-side issue, and retries once on transient failures.

Motivated by a real gpt-5.5 mantle outage (server_error on every request, both regions) that produced opaque failure stickies on core PRs.

• ai_claude-orchestrator.yml: v3.1.4 → v3.1.5 (both jobs)
• ai_claude-backend-reviewer.yml: v3.1.4 → v3.1.5

[github-actions]

PRs linked to this issue

• chore(ai-reviews): bump ai-workflows pin to v3.1.5 by @sfreudenthaler