fix(customer): accurate number_of_purchases type; empty-input percentile test

- find_purchase_percentile's number_of_purchases is now annotated `float`
  (ruff PYI041: float subsumes int) with a docstring noting fractional
  thresholds are valid and compared against the integer counts — matching
  what ensure_number actually accepts. Fractional thresholds like <= 2.5
  are well-defined (equivalent to <= 2), so they are accepted, not rejected.
- Add an empty-input test for DaysBetweenPurchases.purchases_percentile
  (returns NaN), for parity with the PurchasesPerCustomer case.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G93Y8H2PH8oaNNfKssd99f

Author: claude

openretailscience/analysis/customer.py

@@ -153,13 +153,15 @@ def purchases_percentile(self, percentile: float = 0.5) -> float:
 
     def find_purchase_percentile(
         self,
-        number_of_purchases: int,
+        number_of_purchases: float,
         comparison: str = "less_than_equal_to",
     ) -> float:
         """Return the share of customers whose purchase count matches a comparison.
 
         Args:
-            number_of_purchases (int): Threshold to compare against.
+            number_of_purchases (float): Threshold to compare against. Typically an integer count,
+                but fractional values are compared directly against the integer purchase counts
+                (e.g. ``<= 2.5`` is equivalent to ``<= 2``).
             comparison (str): One of ``less_than``, ``less_than_equal_to``,
                 ``equal_to``, ``not_equal_to``, ``greater_than``,
                 ``greater_than_equal_to``. Defaults to ``less_than_equal_to``.

tests/analysis/test_customer.py

@@ -393,6 +393,17 @@ def test_purchases_percentile_uses_linear_interpolation(self, transactions_df, p
         dbp = DaysBetweenPurchases(transactions_df)
         assert dbp.purchases_percentile(percentile) == pytest.approx(expected)
 
+    def test_purchases_percentile_returns_nan_on_empty_input(self):
+        """purchases_percentile returns NaN on empty input (parity with PurchasesPerCustomer)."""
+        empty = pd.DataFrame(
+            {
+                "customer_id": pd.Series([], dtype="int64"),
+                "transaction_date": pd.Series([], dtype="datetime64[ns]"),
+            },
+        )
+        dbp = DaysBetweenPurchases(empty)
+        assert math.isnan(dbp.purchases_percentile(0.5))
+
 
 class TestTransactionChurn:
     """Behavioral tests for TransactionChurn."""