fix(nrinmatriculare): normalizeaza spatii/cratime/puncte consecvent pe toate canalele

Doar 13.8% din numerele de inmatriculare istorice treceau validarea stricta
^[A-Z0-9]{1,10}$ (spike SQL), majoritatea din cauza formatelor reale cu
spatii/cratime ("B 99 XYZ", "B-99-XYZ") pe care normalizarea actuala
(strip+upper) nu le curata.

Adauga normalize_nr_inmatriculare() in app.idempotency (deja punctul comun
API+import pentru canonicalize_row) si o foloseste consecvent in:
- models.PrezentareIn (validare shape / normalizare API)
- validation.validate_prezentare (validare continut)
- idempotency.canonicalize_row (cheia de dedup)

Fara aceasta consecventa, validarea si idempotency ar diverge pe acelasi
numar de inmatriculare in formate echivalente.

app/idempotency.py

@@ -19,9 +19,25 @@ from __future__ import annotations
 
 import hashlib
 import json
+import re
 from typing import Any
 
 
+NRINM_STRIP_RE = re.compile(r"[\s\-.]+")
+
+
+def normalize_nr_inmatriculare(value: object) -> str:
+    """Normalizare canonica nr. inmatriculare: strip, upper, elimina spatii/cratime/puncte.
+
+    Formate reale ("B 99 XYZ", "B-99-XYZ") nu treceau NRINM_RE (app.validation) fara
+    aceasta curatare (spike SQL: doar 13.8% din numerele istorice treceau strict
+    ^[A-Z0-9]{1,10}$). Folosita consecvent in models.PrezentareIn, validation._norm
+    si canonicalize_row de mai jos ca sa nu diverga validarea de cheia de dedup.
+    """
+    s = str(value or "").strip().upper()
+    return NRINM_STRIP_RE.sub("", s)
+
+
 def _op_identity(p: Any) -> str:
     """Cod RAR (normalizat) daca exista, altfel codul intern ROAAUTO."""
     get = p.get if isinstance(p, dict) else (lambda k, d=None: getattr(p, k, d))
@@ -41,7 +57,7 @@ def canonicalize_row(raw: dict[str, Any]) -> dict[str, Any]:
     - prestatii: pastrate ca-atare (rezolvarea e in resolve_prestatii).
     """
     vin = (raw.get("vin") or "").strip().upper()
-    nr = (raw.get("nr_inmatriculare") or "").strip().upper()
+    nr = normalize_nr_inmatriculare(raw.get("nr_inmatriculare"))
 
     # Odometru: strip ".0" Excel float coercion
     odo_raw = raw.get("odometru_final")

app/models.py

@@ -11,6 +11,8 @@ from typing import Literal
 
 from pydantic import BaseModel, Field, field_validator, model_validator
 
+from app.idempotency import normalize_nr_inmatriculare
+
 
 class RarCredentials(BaseModel):
     """Credentiale RAR per-cerere (vin de la ROAAUTO din Oracle). NU se stocheaza."""
@@ -69,11 +71,16 @@ class PrezentareIn(BaseModel):
     obs: str | None = None
     b64_image: str | None = None
 
-    @field_validator("vin", "nr_inmatriculare")
+    @field_validator("vin")
     @classmethod
     def _norm_upper(cls, v: str) -> str:
         return v.strip().upper()
 
+    @field_validator("nr_inmatriculare")
+    @classmethod
+    def _norm_nrinm(cls, v: str) -> str:
+        return normalize_nr_inmatriculare(v)
+
     @field_validator("data_prestatie", "odometru_final")
     @classmethod
     def _norm_strip(cls, v: str) -> str:

app/validation.py

@@ -18,6 +18,7 @@ from datetime import date
 from zoneinfo import ZoneInfo
 
 from app.errors import eroare as _eroare
+from app.idempotency import normalize_nr_inmatriculare
 
 # VIN: 17 caractere, majuscule, fara O/I/Q (plan §2 + contract).
 VIN_RE = re.compile(r"^[A-HJ-NPR-Z0-9]{17}$")
@@ -73,7 +74,7 @@ def validate_prezentare(content: dict) -> list[dict]:
         ))
 
     # --- nrInmatriculare ---
-    nrinm = _norm(content.get("nr_inmatriculare"))
+    nrinm = normalize_nr_inmatriculare(content.get("nr_inmatriculare"))
     if not NRINM_RE.match(nrinm):
         errors.append(_eroare(
             "NR_INMATRICULARE_FORMAT",

tests/test_idempotency.py

@@ -2,7 +2,7 @@
 
 from __future__ import annotations
 
-from app.idempotency import build_key, canonicalize_row
+from app.idempotency import build_key, canonicalize_row, normalize_nr_inmatriculare
 
 
 def _canon():
@@ -26,3 +26,35 @@ def test_key_stabil_pe_env():
     assert build_key(1, canon, "prod") == build_key(1, canon, "prod")
     # None si 1 colapseaza la aceeasi cheie (account_or_default), pe acelasi env
     assert build_key(None, canon, "test") == build_key(1, canon, "test")
+
+
+def test_normalize_nr_inmatriculare_elimina_spatii_cratime_puncte():
+    assert normalize_nr_inmatriculare("B 99 XYZ") == "B99XYZ"
+    assert normalize_nr_inmatriculare("B-99-XYZ") == "B99XYZ"
+    assert normalize_nr_inmatriculare("B.99.XYZ") == "B99XYZ"
+    assert normalize_nr_inmatriculare(" b99xyz ") == "B99XYZ"
+
+
+def test_canonicalize_row_normalizeaza_nr_inmatriculare():
+    raw = {"nr_inmatriculare": "B 99 XYZ"}
+    assert canonicalize_row(raw)["nr_inmatriculare"] == "B99XYZ"
+
+
+def test_build_key_acelasi_pentru_formate_echivalente_nrinm():
+    """"B 99 XYZ" si "B99XYZ" trebuie sa produca aceeasi cheie de dedup."""
+    raw_cu_spatii = {
+        "vin": "WVWZZZ1JZXW000001", "nr_inmatriculare": "B 99 XYZ",
+        "data_prestatie": "2026-01-10", "odometru_final": "123456",
+        "prestatii": [{"cod_prestatie": "OE-1"}],
+    }
+    raw_fara_spatii = {
+        "vin": "WVWZZZ1JZXW000001", "nr_inmatriculare": "B99XYZ",
+        "data_prestatie": "2026-01-10", "odometru_final": "123456",
+        "prestatii": [{"cod_prestatie": "OE-1"}],
+    }
+    canon1 = canonicalize_row(raw_cu_spatii)
+    canon1["prestatii"] = raw_cu_spatii["prestatii"]
+    canon2 = canonicalize_row(raw_fara_spatii)
+    canon2["prestatii"] = raw_fara_spatii["prestatii"]
+    assert build_key(1, canon1, "test") == build_key(1, canon2, "test")
+

tests/test_validation.py

@@ -63,10 +63,28 @@ def test_nrinmatriculare_prea_lung():
 
 
 def test_nrinmatriculare_caracter_special():
-    errors = validate_prezentare(_base(nr_inmatriculare="B-99"))
+    errors = validate_prezentare(_base(nr_inmatriculare="B_99"))
     assert "nr_inmatriculare" in _fields(errors)
 
 
+def test_nrinmatriculare_normalizeaza_spatii():
+    c = _base(nr_inmatriculare="B 99 XYZ")
+    assert c["nr_inmatriculare"] == "B99XYZ"
+    assert validate_prezentare(c) == []
+
+
+def test_nrinmatriculare_normalizeaza_cratime():
+    c = _base(nr_inmatriculare="B-99-XYZ")
+    assert c["nr_inmatriculare"] == "B99XYZ"
+    assert validate_prezentare(c) == []
+
+
+def test_nrinmatriculare_normalizeaza_puncte():
+    c = _base(nr_inmatriculare="B.99.XYZ")
+    assert c["nr_inmatriculare"] == "B99XYZ"
+    assert validate_prezentare(c) == []
+
+
 def test_data_prea_veche():
     errors = validate_prezentare(_base(data_prestatie="2024-11-30"))
     assert "data_prestatie" in _fields(errors)
@@ -135,7 +153,7 @@ def test_b64image_valid_ok():
 
 
 def test_erori_multiple_cumulate():
-    errors = validate_prezentare(_base(vin="BAD", nr_inmatriculare="X-Y", data_prestatie="2024-01-01"))
+    errors = validate_prezentare(_base(vin="BAD", nr_inmatriculare="X_Y", data_prestatie="2024-01-01"))
     assert {"vin", "nr_inmatriculare", "data_prestatie"} <= _fields(errors)
 
 
@@ -215,7 +233,7 @@ def test_back_compat_field_message():
     """Fiecare eroare are inca field + message (forma veche)."""
     errs = validate_prezentare(_base(
         vin="BAD",
-        nr_inmatriculare="X-Y",
+        nr_inmatriculare="X_Y",
         data_prestatie="2024-01-01",
         odometru_final="abc",
         prestatii=[],
@@ -232,7 +250,7 @@ def test_toate_codurile_in_catalog():
 
     cazuri = [
         _base(vin="WVWZZZ1OZAW000123"),                         # VIN_FORMAT
-        _base(nr_inmatriculare="X-Y"),                           # NR_INMATRICULARE_FORMAT
+        _base(nr_inmatriculare="X_Y"),                           # NR_INMATRICULARE_FORMAT
         _base(data_prestatie="15-06-2026"),                      # DATA_FORMAT
         _base(data_prestatie="2024-11-30"),                      # DATA_PREA_VECHE
         _base(data_prestatie=(today_bucuresti() + timedelta(days=1)).isoformat()),  # DATA_VIITOR