Migrate AI auto-review off the sticky_namespace=github.sha hack to use_sticky_comment: false

Context

The ai-automatic-review job in .github/workflows/ai_claude-orchestrator.yml currently forces a fresh review comment per push by setting:

```yaml
sticky_namespace: ${{ github.sha }}
```

This is a workaround that predates first-class support. sticky_namespace was designed to separate different review jobs on the same PR — overloading it with the commit SHA to defeat stickiness is a hack.

Proposal

dotCMS/ai-workflows v3.1.8 (#48 + the #49 fix) adds a supported use_sticky_comment input. Replace the hack with:

```yaml
use_sticky_comment: false
```

Behavior is equivalent (fresh comment per commit, full feedback→change→feedback history) and strictly better:

• The orchestrator scopes the marker by head SHA (the commit actually under review) rather than github.sha (the PR merge commit).
• The marker also includes the model id, so it's collision-safe if a second review model is ever added on the same PR.

Verified e2e on the DeepSeek R1 (bedrock-generic) path in core-workflow-test: in-progress reconciles to final (no orphaned "🔄 in progress" comment), and each commit posts a fresh comment.

Opened as a draft proposal — if there's a reason core specifically wants the sticky_namespace form, we can keep it.

Open .github/workflows/ai_claude-orchestrator.yml and inspect the ai-automatic-review job and its current sticky_namespace setting. Confirm the workflow uses dotCMS/ai-workflows v3.1.8, replace the SHA-based workaround with the supported input, then verify the DeepSeek R1 path in core-workflow-test produces one fresh comment per commit and reconciles in-progress comments to final comments.