feat: stamp the claude CLI version into every eval-store run record

Three harness breakages were traced to claude-CLI TUI drift only after long
flake hunts, because no run record said which CLI it actually exercised.
EvalCollector now stamps claude_cli_version (claude --version, cached once
per process, 'unknown' when the binary is absent) into both partial and
finalized records — schema-additive optional field, no SCHEMA_VERSION bump.
Correlating a flake wave with a CLI release becomes a grep over
~/.gstack/projects/<slug>/evals/ instead of archaeology.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

test/helpers/eval-store.test.ts

@@ -114,6 +114,11 @@ describe('EvalCollector', () => {
     expect(data.total_duration_ms).toBe(3000);
     expect(data.timestamp).toBeTruthy();
     expect(data.hostname).toBeTruthy();
+    // CLI version stamping: always a non-empty string ('unknown' when the
+    // claude binary is absent — the field must exist either way so flake
+    // investigations can correlate runs with the TUI they exercised).
+    expect(typeof data.claude_cli_version).toBe('string');
+    expect(data.claude_cli_version!.length).toBeGreaterThan(0);
   });
 
   test('finalize creates directory if missing', async () => {