chore: update fullsend shim workflow

[fullsend-ai-fullsend]

This PR updates the fullsend shim workflow to match the current template in the .fullsend config repo.

The shim content has drifted from the template — this brings it back in sync.

[github-actions]

Site preview

Preview: https://d0dcdad5-site.fullsend-ai.workers.dev

Commit: 4a92eaacb3e4213c8681d7d1e419fa3d6ae830f2

[fullsend-ai-review]

🤖 Finished Review · ❌ Failure · Started 6:48 AM UTC · Completed 6:48 AM UTC
Commit: 9bd7234 · View workflow run →

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 8:42 AM UTC · Completed 8:47 AM UTC
Commit: ba204cb · View workflow run →

[fullsend-ai-retro]

Retro: PR #2052 — chore: update fullsend shim workflow

This was a trivial 2-line automated onboard PR by fullsend-ai-fullsend[bot] that added a YAML document marker and managed-content comment to .github/workflows/fullsend.yaml. The review agent failed due to authentication — Claude Code CLI reported apiKeySource: "none" and exited immediately with authentication_failed. The harness retried once with the identical result. A human (rh-hemartin) manually approved and merged ~2 hours later.

Existing issues that cover most of the waste here

• #1371 — Skip review dispatch for bot-authored PRs (would have prevented the review run entirely)
• #1630 — Skip retro dispatch for bot-authored PRs with review-only agent interaction (would have prevented this retro)
• #1032 — Pre-flight Vertex AI auth check (would catch auth issues before spinning up the sandbox)

One new proposal filed

The harness should classify agent exit reasons and not retry on authentication failures, which are infrastructure issues that won't self-resolve within the same run.

Proposals filed

• Harness should fail fast on authentication errors instead of retrying (in fullsend-ai/fullsend)