Add evaluation module: RelevancyEvaluator and FactCheckingEvaluator, golden dataset with a pass-rate gate, deterministic CI judge and simulated judge noise

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01JXVi2GMQ7bR5EmbUFdDj7N

evaluation/scripts/run-all.sh

@@ -0,0 +1,16 @@
+#!/usr/bin/env bash
+# Regenerates every file under output/. Tests 01-07 are written by the suite itself through the
+# Transcript helper; 08 is cut from the surefire report to show the live-judge test being skipped.
+# No Docker, no database and no API key is needed.
+set -euo pipefail
+cd "$(dirname "$0")/.."
+rm -rf target
+mvn -q -B test 2>&1 | grep -E "Tests run:|BUILD|FAIL" || true
+f=target/surefire-reports/TEST-com.ankurm.evaluation.LiveJudgeTest.xml
+{
+  echo "# LiveJudgeTest as recorded by surefire when OPENAI_API_KEY is not set"
+  echo
+  grep -o '<testcase name="[^"]*"' "$f" | sed 's/<testcase name=/testcase: /'
+  grep -o '<skipped message="[^"]*"' "$f" | sed 's/<skipped message=/skipped : /'
+} > output/08-live-judge-skipped.txt
+ls output