feat: render the unset-baseline callout on the PR comment

[veksen]

Renderer half of Site #3297. Closes Site #3312. Part of the unset-baseline surfacing thread (Site #3292 follow-ups, epic Site #3106).

Problem

When a project has no comparison branch configured, the baseline silently falls back — to head on a re-push, or (since #147) the PR base. So hasComparison is true and the PR comment renders counts with no warning, e.g. 0 regressed · 0 improved · 0 new on a PR full of new queries (Site #3292, d2armory-next PR #775). The existing "no baseline found / add a push trigger" copy only fires when there's no baseline run at all — a different condition — so this misconfiguration shipped silently.

Change

Site now emits metadata.baseline on POST /ci/runs (Site #3297, merged). This renders it:

• site-api.ts — type the baseline field on CiRunMetadata (optional → tolerates an older Site API and a degraded null read; both render nothing).
• success.md.j2 — a ⚠️ callout when hasComparison and runMetadata.baseline.unset. It names the fallback branch, distinguishes the acute head-vs-head case ("0 new on a PR full of new queries") from a base-branch fallback ("diverges from the dashboard / breaks on non-PR runs"), and surfaces the get_repo_config MCP call to inspect/fix it.

Driven off the Site signal (defined once, Site #3302) rather than re-deriving the rule here.

Scope note

Gated to hasComparison — it targets exactly the gap (a comparison was produced but the baseline is misconfigured). The no-comparison path keeps its existing "add a push trigger" guidance, so the two messages never stack.

Tests

github.test.ts (+4): unset+head-vs-head renders the callout with branch + 0 new + MCP call; base-branch fallback frames the divergence/non-PR risk; configured → no callout; absent/null baseline → no callout. 55/55 reporter tests pass, existing snapshots unchanged, tsc --noEmit clean.

Cross-repo

Producer: Site #3297 (merged) · shared detection Site #3302 · MCP sibling Site #3298 · dashboard sibling Site #3299.

🤖 Generated with Claude Code

[github-actions]

Query Doctor Analysis

3 queries analyzed
0 regressed · 0 improved · 0 new · 0 removed

2 pre-existing issues

• SELECT "guests"."id", "guests"."session_id", "guests"."username", "guests"."avatar_path", "guests"."color", "guests"."side", "guests"."audio_recording_path", "guests"."audio_recording_public", "gue...
  index assets(event_id, inserted_at desc)
  cost 31,003,449 → 1,498 (100% reduction)
• SELECT * FROM guest_ip_addresses WHERE ip_address = '127.0.0.1';
  index guest_ip_addresses(ip_address)
  cost 154,402 → 8 (100% reduction)

_{Using assumed statistics (10000000 rows/table). For better results, sync production stats.}

_{More detail → get_ci_run({ runId: "019ed207-da2d-7965-9d92-72c58c082324" }) · view run · docs}