refactor(chat): AI-native prompts, remove interview heuristics

- Drop interview_reply_length and utterance_substance; always run stage LLM
  and memory retrieval when enabled; trim Settings fields and .env.example.
- Replace guided/opening prompts with compact fact blocks plus unified
  behavior guidance; slim background_voice and persona to tone hints.
- InterviewAgent uses fixed chat_interview max_tokens/chars/segments.

Also includes stacked work: profile followup/extract path, evaluation rubric
and judge schema updates, transcript SPLIT handling in execution service,
user export markdown split tests, and golden case fixture.

api/app/features/evaluation/importers/user_export_markdown.py

@@ -5,6 +5,11 @@ from __future__ import annotations
 import re
 
 
+def _normalize_export_ai_block(body: str) -> str:
+    """多段助手回复中的 [SPLIT] 在 Markdown 导出中改为换行，便于阅读。"""
+    return (body or "").replace("[SPLIT]", "\n").strip()
+
+
 def extract_user_utterances_from_export_md(text: str) -> list[str]:
     """匹配 ``**用户:**`` 块之间的正文。"""
     out: list[str] = []
@@ -41,6 +46,6 @@ def extract_dialogue_turns_from_export_md(text: str) -> list[tuple[str, str]]:
         u = (user_m.group(1) or "").strip()
         if not u or u == "（空）":
             continue
-        a = ((ai_m.group(1) if ai_m else "") or "").strip()
-        out.append((u, a))
+        raw_ai = ((ai_m.group(1) if ai_m else "") or "").strip()
+        out.append((u, _normalize_export_ai_block(raw_ai)))
     return out