feat(embeddings): cache persistent de vectori in SQLite + warmup in fundal

Vectorii corpusului k-NN persista in tabela embedding_cache (PK model+text_hash,
blob float32 LE); la warmup se vectorizeaza doar textele lipsa din cache, deci
restartul cu corpus neschimbat nu mai plateste ~1-2 min de embed (embed=0).

- app/embedding_cache.py: serializare array('f'), load/save/purge chunk 500 cu
  BEGIN/COMMIT explicit (conexiuni autocommit), validare dimensiune la scriere
  si citire, orchestrare sync_corpus_vectors cu embed_fn injectat
- index_corpus(vectors=): vectori precalculati cu validare aliniere; mismatch
  -> fallback embed complet
- ensure_embeddings_corpus: warmup in thread la startup (block=True), calea de
  request ne-blocanta (acquire non-blocking pe lock; warmup in curs -> return
  imediat); purjare orfane + modele vechi doar dupa indexare reusita
- log warmup: cache=N embed=M in Xs
- 31 teste noi (cold/warm/incremental, model schimbat, concurenta, ranking
  exact, echivalenta float32); suita completa 1596 passed

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

tests/test_embeddings.py

@@ -166,6 +166,79 @@ def test_index_corpus_no_exception_on_backend_error():
     assert engine.suggest_nearest("CEVA") == []
 
 
+# --------------------------------------------------------------------------- #
+# index_corpus(vectors=) -- precalculati, aliniati cu items (A8)             #
+# --------------------------------------------------------------------------- #
+
+def test_index_corpus_vectors_precalculati_nu_apeleaza_backend():
+    """Cand `vectors` e furnizat, backend-ul NU e apelat pentru corpus (embed=0)."""
+
+    class NoCallBackend:
+        def embed(self, texts):
+            raise AssertionError("backend.embed() NU trebuie apelat cand vectors e furnizat")
+
+    corpus = [
+        {"denumire": "SCHIMB ULEI", "cod": "OE-3"},
+        {"denumire": "REPARATIE MOTOR", "cod": "OE-1"},
+    ]
+    vectors = [_vec("SCHIMB ULEI"), _vec("REPARATIE MOTOR")]
+    engine = EmbeddingEngine(backend=NoCallBackend())
+    engine.index_corpus(corpus, vectors=vectors)
+    assert engine.has_corpus()
+
+
+def test_index_corpus_vectors_ranking_exact_warm_start():
+    """Vectori precalculati din 'cache' produc EXACT acelasi cod ca embed direct
+    (nu doar non-empty) -- prinde o eventuala dezaliniere intre items si vectors."""
+    corpus = [
+        {"denumire": "SCHIMB ULEI MOTOR", "cod": "OE-3"},
+        {"denumire": "REPARATIE CUTIE VITEZE", "cod": "OE-1"},
+        {"denumire": "VERIFICARE DIRECTIE VOLAN", "cod": "OE-4"},
+        {"denumire": "INLOCUIT PLACUTE FRANA", "cod": "OE-2"},
+    ]
+    vectors = [_vec(item["denumire"]) for item in corpus]
+    engine = EmbeddingEngine(backend=MockBackend())
+    engine.index_corpus(corpus, vectors=vectors)
+
+    assert engine.suggest_nearest("SCHIMB ULEI MOTOR", top_k=1)[0]["cod"] == "OE-3"
+    assert engine.suggest_nearest("VERIFICARE DIRECTIE VOLAN", top_k=1)[0]["cod"] == "OE-4"
+
+
+def test_index_corpus_vectors_mismatch_lungime_fallback_embed_complet():
+    """len(vectors) != len(items) -> fallback pe embed complet (backend chemat), fara exceptie."""
+    corpus = [
+        {"denumire": "SCHIMB ULEI", "cod": "OE-3"},
+        {"denumire": "REPARATIE MOTOR", "cod": "OE-1"},
+    ]
+    engine = EmbeddingEngine(backend=MockBackend())
+    engine.index_corpus(corpus, vectors=[_vec("SCHIMB ULEI")])  # un singur vector pentru 2 itemi
+
+    assert engine.has_corpus()
+    results = engine.suggest_nearest("SCHIMB ULEI", top_k=1)
+    assert results and results[0]["cod"] == "OE-3"
+
+
+def test_index_corpus_vectors_contine_none_fallback_embed_complet():
+    """Un `None` in `vectors` -> fallback pe embed complet, fara exceptie."""
+    corpus = [
+        {"denumire": "SCHIMB ULEI", "cod": "OE-3"},
+        {"denumire": "REPARATIE MOTOR", "cod": "OE-1"},
+    ]
+    engine = EmbeddingEngine(backend=MockBackend())
+    engine.index_corpus(corpus, vectors=[_vec("SCHIMB ULEI"), None])
+
+    assert engine.has_corpus()
+    assert engine.suggest_nearest("SCHIMB ULEI", top_k=1)[0]["cod"] == "OE-3"
+
+
+def test_index_corpus_vectors_none_e_comportamentul_existent():
+    """`vectors=None` (default) -> embed complet prin backend, ca inainte."""
+    corpus = [{"denumire": "SCHIMB ULEI", "cod": "OE-3"}]
+    engine = EmbeddingEngine(backend=MockBackend())
+    engine.index_corpus(corpus, vectors=None)
+    assert engine.suggest_nearest("SCHIMB ULEI", top_k=1)[0]["cod"] == "OE-3"
+
+
 # --------------------------------------------------------------------------- #
 # API la nivel de modul (singleton global)                                   #
 # --------------------------------------------------------------------------- #