feat(api): 收敛对话与记忆流程边界，引入 LLM 网关与专用服务

- MemoryService 异步路径委托 MemoryIngestService / MemoryRetrievalService；富化派发经 MemoryEnrichmentScheduler
- WebSocket pipeline 经 ChatTurnService 与显式 DTO 编排单轮对话；回忆录片段入队由 MemoirIngestScheduler 封装
- 新增 LlmGateway（LlmUseCase），各 agent、任务与适配器对齐 ports
- 补充 memory 提示适配、runtime 类型、memory-retrieval 文档、ai-touchpoints 说明与扫描脚本及配套测试

Made-with: Cursor

api/tests/test_memory_evidence.py

@@ -2,6 +2,7 @@
 
 import pytest
 
+from app.core.config import settings
 from app.features.memory import evidence as evidence_mod
 from app.features.memory.evidence_format import format_evidence_chunks_for_chat_prompt
 from app.features.memory.evidence import (
@@ -9,6 +10,7 @@ from app.features.memory.evidence import (
     _facts_to_dicts,
     _stories_to_dicts,
     _timeline_to_dicts,
+    retrieve_evidence_bundle_async,
     retrieve_evidence_bundle_sync,
 )
 from app.features.memory.schemas import EvidenceBundle
@@ -190,3 +192,69 @@ def test_slice_interview_memory_suppresses_long_new_topic():
     s = slice_interview_memory(evidence, long_msg)
     assert s.prompt_excerpt == ""
     assert s.anchor_source == ""
+
+
+async def test_retrieve_evidence_bundle_async_non_empty_merges_precomputed_chunks(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """非空 query：异步路径以 merged_chunk_dicts 为主，元数据来自 fetch_evidence_metadata_async。"""
+    meta = {
+        "relevant_facts": [
+            {
+                "id": "f1",
+                "fact_type": "bio",
+                "subject": "s",
+                "predicate": "p",
+                "object_json": {},
+            }
+        ],
+        "timeline_hints": [],
+        "relevant_summaries": [
+            {
+                "id": "s1",
+                "summary_type": "session",
+                "content": "sum",
+                "source_chunk_ids": [],
+            }
+        ],
+        "relevant_stories": [],
+    }
+
+    async def fake_fetch_meta(db, user_id, q, top_k):
+        assert user_id == "u1"
+        assert q == "hello"
+        assert top_k == 7
+        return meta
+
+    monkeypatch.setattr(evidence_mod, "fetch_evidence_metadata_async", fake_fetch_meta)
+    merged = [{"id": "c1", "content": "chunk body", "chunk_index": 0}]
+    out = await retrieve_evidence_bundle_async(
+        object(),
+        "u1",
+        "  hello  ",
+        top_k=7,
+        merged_chunk_dicts=merged,
+    )
+    assert out == {"relevant_chunks": merged, **meta}
+
+
+async def test_empty_query_evidence_bundle_async_and_sync_aligned_when_rolling_off(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(settings, "memory_evidence_empty_query_include_rolling", False)
+    out_a = await retrieve_evidence_bundle_async(
+        object(),
+        "u1",
+        "   ",
+        top_k=10,
+        merged_chunk_dicts=[],
+    )
+    assert out_a == dict(EMPTY_EVIDENCE_BUNDLE)
+    out_s = retrieve_evidence_bundle_sync(
+        session=object(),
+        user_id="u1",
+        query="",
+        top_k=10,
+        embedding_provider=None,
+    )
+    assert out_s == dict(EMPTY_EVIDENCE_BUNDLE)