audit-4: classifier understands Spanish negations (veto unchanged)

Merchants write in Spanish; negations like nunca, lamentablemente or
no podemos must veto approval exactly like their English counterparts.
Previously an all-Spanish rejection mail could fall through to OTHER,
and a mixed mail with the approval word in English approved despite a
Spanish negation.

- negation list adds: nunca, jamas, tampoco, ningun/ninguna, nadie,
  nada, no podemos, no es posible, no aceptamos, rechaz*, descartad*,
  lamentablemente, no ha sido aprobad*, no procede
- new Spanish REJECTION rule (rechaz*, solicitud denegada, no cumple
  los requisitos) and Spanish APPROVAL rule (aprob*, aceptad*,
  bienvenid*, encantados de trabajar)
- veto rule unchanged: any negation anywhere -> needs_review
- tests: 15 new (12 veto cases incl. 2 mixed-language, pure-Spanish
  approval still approves, Spanish rejection classifies rejection)

Gate: ruff 0 · 96 unit + 1 PG integration · gitleaks clean.

tests/test_spanish_negations.py

@@ -0,0 +1,80 @@
+"""Spanish negation tests (round-3 point 4): merchants write in Spanish;
+negations like 'nunca', 'lamentablemente' or 'no podemos' must veto
+`approval` exactly like their English counterparts. Veto rule unchanged:
+when in doubt -> needs_review."""
+
+from __future__ import annotations
+
+import pytest
+
+from cayceops.components.maild import classify as clf
+from cayceops.models import MailClassification
+
+APPROVAL_ONLY = MailClassification.APPROVAL
+
+
+def _never_approval(cfg_factory, text: str) -> None:
+    kind, _ = clf.classify(text, cfg_factory())
+    assert kind is not APPROVAL_ONLY, f"{text!r} classified as APPROVAL"
+
+
+def test_english_baseline_still_works(cfg):
+    config = cfg()
+    kind, _ = clf.classify("Your application was approved, welcome to the program", config)
+    assert kind is APPROVAL_ONLY
+
+
+@pytest.mark.parametrize(
+    "text",
+    [
+        # 1. rejection stem
+        "Lamentamos informarle que su solicitud fue rechazada",
+        # 2. nunca
+        "Nunca aceptamos negocios de criptomonedas",
+        # 3. jamás
+        "Jamás hemos trabajado con affiliates externos",
+        # 4. tampoco
+        "No podemos aceptar su sitio, tampoco podemos ofrecerle otro acuerdo",
+        # 5. ningún/ninguna
+        "En este momento no tenemos ningún hueco para nuevos partners",
+        # 6. no podemos (multi-word)
+        "Su propuesta es interesante pero no podemos proceder",
+        # 7. no es posible (multi-word)
+        "Gracias por escribir, pero no es posible colaborar este trimestre",
+        # 8. lamentablemente
+        "Lamentablemente su aplicación no ha sido aprobada",
+        # 9. no aceptamos
+        "Por política interna no aceptamos sitios de viajes",
+        # 10. mixed spanish/english: negation in spanish, approval word in english
+        "Thanks for reaching out! Lamentablemente we cannot approve your site",
+        # 11. mixed: approval in spanish, negation in english
+        "Apreciamos su interés, but we are not accepting new partners",
+        # 12. nadie / nada
+        "Aquí nadie decidiría eso, así que no hay nada que aprobar",
+    ],
+)
+def test_spanish_and_mixed_negations_veto_approval(cfg, text):
+    _never_approval(cfg, text)
+
+
+def test_spanish_pure_approval_still_approves(cfg):
+    """A clean Spanish approval WITHOUT any negation must still classify
+    approval — the veto must not over-fire."""
+    config = cfg()
+    kind, confidence = clf.classify(
+        "Su solicitud fue aprobada. Bienvenido al programa, estamos encantados "
+        "de trabajar con ustedes.",
+        config,
+    )
+    assert kind is APPROVAL_ONLY
+    assert confidence >= 0.5
+
+
+def test_spanish_rejection_classifies_rejection(cfg):
+    config = cfg()
+    kind, conf = clf.classify(
+        "Lamentamos informarle que su solicitud fue rechazada por el comité",
+        config,
+    )
+    assert kind is MailClassification.REJECTION
+    assert conf >= 0.9