Remove vision OCR — use Tesseract-only pipeline for receipt parsing

The llama3.2-vision model was producing unreliable structured data
(wrong vendors, amounts, dates) making expense reports worse than
Tesseract + LLM extraction.  Removes _ocr_image_vision(), the
vision JSON fast path in _parse_receipt_text(), _match_category(),
and the vision_ocr_model config setting entirely.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

tests/test_expenses_agent.py

@@ -423,88 +423,12 @@ async def test_act_no_employee_returns_empty_and_escalates():
 
 
 # ---------------------------------------------------------------------------
-# _match_category
-# ---------------------------------------------------------------------------
-
-class TestMatchCategory:
-    PRODUCTS = [
-        {'id': 1, 'name': 'Meals'},
-        {'id': 2, 'name': 'Fuel'},
-        {'id': 3, 'name': 'Hotel'},
-        {'id': 4, 'name': 'Office Supplies'},
-        {'id': 5, 'name': 'Transport'},
-        {'id': 6, 'name': 'Other'},
-    ]
-
-    def test_exact_match(self):
-        assert ExpensesAgent._match_category('Meals', self.PRODUCTS) == 'Meals'
-
-    def test_case_insensitive(self):
-        assert ExpensesAgent._match_category('meals', self.PRODUCTS) == 'Meals'
-        assert ExpensesAgent._match_category('FUEL', self.PRODUCTS) == 'Fuel'
-
-    def test_substring_match(self):
-        # 'office' is a substring of 'Office Supplies'
-        assert ExpensesAgent._match_category('office', self.PRODUCTS) == 'Office Supplies'
-
-    def test_fuzzy_match(self):
-        # 'transport' is close to 'Transport'
-        assert ExpensesAgent._match_category('transport', self.PRODUCTS) == 'Transport'
-
-    def test_no_match_returns_empty(self):
-        assert ExpensesAgent._match_category('zxqwerty', self.PRODUCTS) == ''
-
-    def test_empty_category(self):
-        assert ExpensesAgent._match_category('', self.PRODUCTS) == ''
-
-    def test_empty_products(self):
-        assert ExpensesAgent._match_category('meals', []) == ''
-
-
-# ---------------------------------------------------------------------------
-# _parse_receipt_text — vision JSON fast path
+# _parse_receipt_text — LLM extraction path
 # ---------------------------------------------------------------------------
 
 @pytest.mark.asyncio
-async def test_parse_vision_json_fast_path():
-    """When text is pre-extracted JSON from vision model, skip LLM call."""
-    agent = _make_agent()
-    agent._llm.submit = AsyncMock()  # should NOT be called
-
-    vision_json = ('{"vendor":"McDonald\'s","amount":12.50,'
-                   '"date":"2026-05-09","time":"13:30","category":"meals"}')
-    products = [{'id': 1, 'name': 'Meals'}, {'id': 2, 'name': 'Fuel'}]
-
-    result = await agent._parse_receipt_text(vision_json, 'receipt.jpg',
-                                              expense_products=products)
-
-    assert result['vendor'] == "McDonald's"
-    assert result['amount'] == 12.50
-    assert result['date'] == '2026-05-09'
-    assert result['time'] == '13:30'
-    assert result['product_name'] == 'Meals'
-    agent._llm.submit.assert_not_called()
-
-
-@pytest.mark.asyncio
-async def test_parse_vision_json_null_time():
-    """Vision model may return the string 'null' for time — normalise to None."""
-    agent = _make_agent()
-    agent._llm.submit = AsyncMock()
-
-    vision_json = '{"vendor":"Shell","amount":45.00,"date":"2026-05-09","time":"null","category":"fuel"}'
-    products = [{'id': 1, 'name': 'Meals'}, {'id': 2, 'name': 'Fuel'}]
-
-    result = await agent._parse_receipt_text(vision_json, 'shell.jpg',
-                                              expense_products=products)
-    assert result['time'] is None
-    assert result['product_name'] == 'Fuel'
-    agent._llm.submit.assert_not_called()
-
-
-@pytest.mark.asyncio
-async def test_parse_non_json_text_falls_through_to_llm():
-    """Plain OCR text (not JSON) should go through the LLM extraction path."""
+async def test_parse_plain_ocr_text_uses_llm():
+    """Plain OCR text should go through the LLM extraction path."""
     agent = _make_agent()
     llm_resp = MagicMock()
     llm_resp.content = '{"vendor":"Acme","amount":9.99,"date":"2026-05-09","time":null,"product_name":"Meals"}'