feat(search): Phase 2A E-1 — Qwen 후보 3종 백필 CLI + eval 디스패처 확장 (마이그 328~333)

- 후보 섀도 테이블 6종(전부 vector 타입 — eval=exact scan 이라 인덱스 불요, halfvec 은 C-1 소관)
- workers/phase2a_cand_backfill: resumable(NOT EXISTS)·배치 커밋·동결셋 한정(--doc/chunk-id-max),
  문서/청크 입력 = production 경로 동일 구성 + plain
- CANDIDATE_BACKEND_MAP += cand_qwen06/qwen4/qwen4m (embed_kind=ollama, 쿼리측 instruct prefix
  G-1 핀 문자열, qwen4m = dimensions 1024 MRL)
- qwen4m 적재는 qwen4 에서 SQL 파생(subvector+l2_normalize) — 본 CLI 비대상

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

tests/test_phase2a_backend.py

@@ -0,0 +1,96 @@
+"""Phase 2A (embedding-phase2a-1) — Qwen 후보 디스패처/쿼리 임베딩 단위 테스트."""
+
+from __future__ import annotations
+
+import pytest
+
+from services.search import retrieval_service as rs
+
+
+def test_resolve_qwen_backends():
+    for slug in ("cand_qwen06", "cand_qwen4", "cand_qwen4m"):
+        cfg = rs._resolve_backend(slug)
+        assert cfg["docs_table"].startswith("documents_cand_qwen")
+        assert cfg["chunks_table"].startswith("document_chunks_cand_qwen")
+        assert cfg["embed_kind"] == "ollama"
+        # 테이블명이 2단계 SQL allowlist 도 통과해야 함 (R2-B1)
+        assert rs._VALID_DOCS_TABLE.match(cfg["docs_table"])
+        assert rs._VALID_CHUNKS_TABLE.match(cfg["chunks_table"])
+    assert rs._resolve_backend("baseline") is None
+    with pytest.raises(ValueError):
+        rs._resolve_backend("cand_unknown")
+
+
+def test_qwen4m_has_mrl_dimensions():
+    assert rs._resolve_backend("cand_qwen4m")["embed_dimensions"] == 1024
+    assert "embed_dimensions" not in rs._resolve_backend("cand_qwen4")
+
+
+class _FakeResp:
+    def __init__(self, embs):
+        self._embs = embs
+
+    def raise_for_status(self):
+        return None
+
+    def json(self):
+        return {"embeddings": self._embs}
+
+
+class _FakeClient:
+    """httpx.AsyncClient 대역 — post body 캡처."""
+
+    captured: dict = {}
+
+    def __init__(self, *a, **k):
+        pass
+
+    async def __aenter__(self):
+        return self
+
+    async def __aexit__(self, *a):
+        return False
+
+    async def post(self, url, json=None):
+        _FakeClient.captured = {"url": url, "json": json}
+        dim = (json or {}).get("dimensions") or 1024
+        return _FakeResp([[0.1] * dim])
+
+
+@pytest.mark.asyncio
+async def test_ollama_query_embed_applies_instruct_prefix(monkeypatch):
+    import httpx
+
+    monkeypatch.setattr(httpx, "AsyncClient", _FakeClient)
+    cfg = rs._resolve_backend("cand_qwen06")
+    out = await rs._embed_query_via_ollama(cfg, "압력용기 수압시험")
+    assert out is not None and len(out) == 1024
+    body = _FakeClient.captured["json"]
+    assert body["model"] == "qwen3-embedding:0.6b"
+    assert body["input"][0].startswith(rs.QWEN3_QUERY_INSTRUCT)
+    assert body["input"][0].endswith("압력용기 수압시험")
+    assert "dimensions" not in body
+
+
+@pytest.mark.asyncio
+async def test_ollama_query_embed_mrl_dimensions(monkeypatch):
+    import httpx
+
+    monkeypatch.setattr(httpx, "AsyncClient", _FakeClient)
+    cfg = rs._resolve_backend("cand_qwen4m")
+    out = await rs._embed_query_via_ollama(cfg, "q")
+    assert _FakeClient.captured["json"]["dimensions"] == 1024
+    assert len(out) == 1024
+
+
+@pytest.mark.asyncio
+async def test_ollama_query_embed_failure_returns_none(monkeypatch):
+    import httpx
+
+    class _Boom(_FakeClient):
+        async def post(self, url, json=None):
+            raise httpx.ConnectError("down")
+
+    monkeypatch.setattr(httpx, "AsyncClient", _Boom)
+    cfg = rs._resolve_backend("cand_qwen06")
+    assert await rs._embed_query_via_ollama(cfg, "q") is None