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/.env.example

@@ -53,11 +53,10 @@ EMBEDDING_MODEL=embedding-3
 # CHAT_STAGE_DETECTION_MAX_TOKENS=128
 # 访谈性格（InterviewAgent）：default | warm_listener | curious_guide
 # CHAT_INTERVIEW_PERSONA=default
-# 访谈回复长度档位（brief/standard/expanded）联动：极短输入 / 默认 / 长段+新细节
-# CHAT_INTERVIEW_BRIEF_MAX_TOKENS=240
-# CHAT_INTERVIEW_BRIEF_MAX_CHARS_PER_SEGMENT=180
-# CHAT_INTERVIEW_EXPANDED_MAX_TOKENS=400
-# CHAT_INTERVIEW_EXPANDED_MAX_CHARS_PER_SEGMENT=300
+# 访谈主回复：统一 max_tokens / 单段字数（代码截断），不再分 brief/expanded 档
+# CHAT_INTERVIEW_MAX_TOKENS=380
+# CHAT_INTERVIEW_MAX_CHARS_PER_SEGMENT=260
+# CHAT_INTERVIEW_MAX_SEGMENTS=2
 # 访谈：是否按本轮用户话检索记忆并注入提示词（关则不调 retrieve）
 # CHAT_MEMORY_RETRIEVAL_ENABLED=true
 # CHAT_MEMORY_TOP_K=8
@@ -82,13 +81,6 @@ EMBEDDING_MODEL=embedding-3
 # CHAT_INPUT_NORMALIZE_LLM_MAX_INPUT_CHARS=8000
 # True：仅 is_from_voice 时走 LLM 纠错；键盘输入仅规则归一
 # CHAT_INPUT_NORMALIZE_LLM_VOICE_ONLY=true
-# 短时/应答/元话语：本轮跳过阶段 LLM 与记忆向量检索（仍保留访谈主 LLM）；关则每轮完整路径
-# CHAT_SUBSTANTIVE_HEURISTIC_ENABLED=true
-# CHAT_SUBSTANTIVE_MIN_CHARS=12
-# CHAT_STAGE_DETECTION_SKIP_LLM_ON_INSUFFICIENT_SIGNAL=true
-# CHAT_MEMORY_RETRIEVAL_REQUIRE_SUBSTANTIVE=true
-# 资料收集：短时/元话语不跑资料字段抽取 LLM（仍生成追问）
-# CHAT_PROFILE_EXTRACT_REQUIRE_SUBSTANTIVE=true
 
 # Memoir Phase1：True 时用一次「批量 JSON」做抽取+分类（单段或多段均可；失败自动回退逐段）。
 # False 时始终逐段（与启用本开关前的行为一致，含防抖合并后的多段任务）。