fix(bench): init OpenCode client before scoring baseline

`scoreBaseline()` called `matchTruthFinding()` which invokes
`sendPrompt()` — but unlike `runScorer()`, `scoreBaseline()` never
called `initOpenCodeClient()` first. Result: every run exited with

  [bench] ERROR: OpenCode client not initialized. Call initOpenCodeClient() first.

regardless of whether baseline and truth data were present.

Call `initOpenCodeClient(config.model)` at the top of `scoreBaseline()`
so the two scoring paths have equivalent init behavior.

Author: VoidChecksum

packages/bench/src/scorer/baseline-scorer.ts

@@ -1,6 +1,7 @@
 import type { ScaBenchBaseline, ScoringMetadata, VigiloFinding, ScorerMatch } from "../types.js";
 import type { ScorerConfig } from "../utils.js";
 import { matchTruthFinding } from "./llm-scorer.js";
+import { initOpenCodeClient } from "../client/opencode.js";
 import { log } from "../utils.js";
 import pc from "picocolors";
 
@@ -57,6 +58,11 @@ export async function scoreBaseline(
     log(pc.dim(`Truth findings: ${truthFindings.length}`));
   }
 
+  // runScorer() initializes the OpenCode client; scoreBaseline() skipped it
+  // historically, which surfaced only as "client not initialized" on first
+  // sendPrompt(). Initialize explicitly so the two paths behave the same.
+  await initOpenCodeClient(config.model);
+
   // Convert baseline findings to VigiloFinding format
   const workingSet: WorkingFinding[] = baseline.findings.map((f, idx) => ({
     id: f.id,