Fix float coercion for numeric filter parameters (#2876)

Closes #1681

tests/test_numeric_filter_values.py

@@ -0,0 +1,45 @@
+import sqlite3
+
+import pytest
+
+from datasette.filters import Filters
+
+
+@pytest.mark.parametrize(
+    "value,expected",
+    (
+        ("3.5", 3.5),
+        ("-2", -2),
+        ("-2.5", -2.5),
+        ("1e3", 1000.0),
+        ("not-a-number", "not-a-number"),
+        ("nan", "nan"),
+        ("inf", "inf"),
+        ("-inf", "-inf"),
+    ),
+)
+def test_numeric_filter_parameters(value, expected):
+    filters = Filters((("score__gt", value),))
+    sql_bits, params = filters.build_where_clauses("items")
+
+    assert sql_bits == ['"score" > :p0']
+    assert params == {"p0": expected}
+
+
+def test_numeric_filter_parameters_against_calculated_view():
+    conn = sqlite3.connect(":memory:")
+    conn.execute("create table searchable(pk integer)")
+    conn.executemany("insert into searchable(pk) values (?)", [(0,), (1,), (2,)])
+    conn.execute(
+        "create view calculated as "
+        "select pk + 1 as pk_plus_one, pk / 2.0 as score from searchable"
+    )
+
+    sql_bits, params = Filters((("score__gt", "0.1"),)).build_where_clauses(
+        "calculated"
+    )
+    rows = conn.execute(
+        "select score from calculated where {}".format(" and ".join(sql_bits)), params
+    ).fetchall()
+
+    assert rows == [(0.5,), (1.0,)]