Fix optimize crash: normalize LLM strategies that return dicts instead of strings

The analysis LLM sometimes returns strategies as objects like
[{"name": "clarity"}] instead of plain strings ["clarity"]. The
', '.join(strategies) call then fails with "expected str, got dict".

Fix: normalize each strategy to a string regardless of format —
handles str, dict with name/strategy keys, or fallback to str().

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

llm_team_ui.py

@@ -7531,7 +7531,18 @@ def _run_optimize(job_id, run_id):
             j_e = analysis_raw.rfind("}") + 1
             if j_s >= 0 and j_e > j_s:
                 parsed = json.loads(analysis_raw[j_s:j_e])
-                strategies = parsed.get("strategies", strategies)[:5]
+                raw_strats = parsed.get("strategies", strategies)[:5]
+                # Normalize — LLM might return strings or dicts
+                strategies = []
+                for s in raw_strats:
+                    if isinstance(s, str):
+                        strategies.append(s)
+                    elif isinstance(s, dict):
+                        strategies.append(s.get("name", s.get("strategy", str(s))))
+                    else:
+                        strategies.append(str(s))
+                if not strategies:
+                    strategies = ["clarity", "depth", "specificity"]
         except Exception:
             pass