Fix classify_test: use last record's offload status for failed tests

For failed tests, failure always occurs at the last record. The previous
logic used any(FALLBACK) across all records, misclassifying tests where
intermediate records were FALLBACK but the failing last record was OFFLOAD.

Co-Authored-By: Claude Opus 4 <noreply@anthropic.com>

Author: baibaichen

.github/scripts/analyze-ansi.py

@@ -205,24 +205,29 @@ def _infer_job_name(xml_path):
 # ANALYSIS LAYER
 # ===========================================================================
 
-def classify_test(status, has_fallback, has_offload_data):
-    """Unified three-color classification."""
+def classify_test(status, has_fallback, has_offload_data, last_record_fallback=None):
+    """Unified three-color classification.
+
+    For failed tests, last_record_fallback (bool) determines Offload vs Fallback
+    because failure always occurs at the last record.
+    """
     is_pass = status in ("PASSED", "PASS")
     is_skip = status in ("SKIPPED", "SKIP")
+    is_fail = status in ("FAILED", "ERROR", "FAIL")
     if is_skip:
         return "⚪", "Skipped"
-    if has_fallback:
-        if is_pass:
-            return "🔴", "Passed+Fallback"
-        return "🔴", "Failed+Fallback"
     if not has_offload_data:
         if is_pass:
             return "⚪", "Passed (no data)"
         return "🟡", "Failed (no data)"
+    if is_fail:
+        if last_record_fallback:
+            return "🔴", "Failed+Fallback"
+        return "🟡", "Failed+Offload"
+    if has_fallback:
+        return "🔴", "Passed+Fallback"
     if is_pass:
         return "🟢", "Passed+Offload"
-    if status in ("FAILED", "ERROR", "FAIL"):
-        return "🟡", "Failed+Offload"
     return "🟡", "Failed"
 
 
@@ -236,8 +241,10 @@ def analyze_json_tests(suites):
             records = t.get("records", [])
             has_fallback = any(r.get("offload") == "FALLBACK" for r in records)
             has_offload_data = len(records) > 0
+            last_record_fallback = (records[-1].get("offload") == "FALLBACK") if records else None
             color, label = classify_test(
-                t.get("status", "PASSED"), has_fallback, has_offload_data)
+                t.get("status", "PASSED"), has_fallback, has_offload_data,
+                last_record_fallback)
             tests.append({
                 "suite": suite_name,
                 "test": t["name"],
@@ -469,8 +476,10 @@ def format_summary(summary, json_tests, suites=None):
                 has_fallback = any(r.get("offload") == "FALLBACK"
                                    for r in records)
                 has_offload_data = len(records) > 0
+                last_fb = (records[-1].get("offload") == "FALLBACK") if records else None
                 _, label = classify_test(
-                    t.get("status", "PASSED"), has_fallback, has_offload_data)
+                    t.get("status", "PASSED"), has_fallback, has_offload_data,
+                    last_fb)
                 counts[label] += 1
             total = sum(counts.values())
             po = counts.get("Passed+Offload", 0)