Fix ValueTracker skipping type evaluation for empty strings

`ValueTracker.evaluate()` used `if not value` to skip NULL values, but
this also skipped empty strings (`""`), `0`, and other falsy values.
A CSV column containing only empty strings was incorrectly inferred as
INTEGER (the initial default) instead of TEXT.

Replace the falsy check with an explicit `if value is None` so that only
NULL is skipped and all other values — including `""` — are evaluated
against the type tests.

sqlite_utils/utils.py

@@ -490,7 +490,7 @@ class ValueTracker:
         return "text"
 
     def evaluate(self, value: object) -> None:
-        if not value or not self.couldbe:
+        if value is None or not self.couldbe:
             return
         not_these: List[str] = []
         for name, test in self.couldbe.items():

tests/test_utils.py

@@ -1,4 +1,5 @@
 from sqlite_utils import utils
+from sqlite_utils.utils import TypeTracker
 import csv
 import io
 import pytest
@@ -73,6 +74,23 @@ def test_maximize_csv_field_size_limit():
     assert rows_list2[0]["text"] == long_value
 
 
+@pytest.mark.parametrize(
+    "rows,expected_types",
+    [
+        ([{"a": ""}], {"a": "text"}),
+        ([{"a": ""}, {"a": ""}], {"a": "text"}),
+        ([{"a": "1"}, {"a": ""}], {"a": "text"}),
+        ([{"a": "0"}], {"a": "integer"}),
+        ([{"a": "1"}], {"a": "integer"}),
+        ([{"a": None}], {"a": "integer"}),
+    ],
+)
+def test_type_tracker_empty_strings(rows, expected_types):
+    tracker = TypeTracker()
+    list(tracker.wrap(iter(rows)))
+    assert tracker.types == expected_types
+
+
 @pytest.mark.parametrize(
     "input,expected",
     (