[breaking] Bump Python requirement to 3.10. (#10434)

- Bump the Python requirement.
- Fix type hints.
- Use loky to avoid deadlock.
- Workaround cupy-numpy compatibility issue on Windows caused by the `safe` casting rule.
- Simplify the repartitioning logic to avoid dask errors.

tests/python-gpu/test_device_quantile_dmatrix.py

@@ -20,7 +20,7 @@ class TestQuantileDMatrix:
     def test_dmatrix_feature_weights(self) -> None:
         import cupy as cp
 
-        rng = cp.random.RandomState(1994)
+        rng = cp.random.RandomState(np.uint64(1994))
         data = rng.randn(5, 5)
         m = xgb.DMatrix(data)
 
@@ -146,7 +146,7 @@ class TestQuantileDMatrix:
     def test_metainfo(self) -> None:
         import cupy as cp
 
-        rng = cp.random.RandomState(1994)
+        rng = cp.random.RandomState(np.uint64(1994))
 
         rows = 10
         cols = 3
@@ -170,7 +170,7 @@ class TestQuantileDMatrix:
     def test_ref_dmatrix(self) -> None:
         import cupy as cp
 
-        rng = cp.random.RandomState(1994)
+        rng = cp.random.RandomState(np.uint64(1994))
         self.cputest.run_ref_dmatrix(rng, "gpu_hist", False)
 
     @given(

tests/python-gpu/test_from_cupy.py

@@ -66,7 +66,7 @@ def _test_from_cupy(DMatrixT):
 
 def _test_cupy_training(DMatrixT):
     np.random.seed(1)
-    cp.random.seed(1)
+    cp.random.seed(np.uint64(1))
     X = cp.random.randn(50, 10, dtype="float32")
     y = cp.random.randn(50, dtype="float32")
     weights = np.random.random(50) + 1
@@ -131,7 +131,7 @@ def _test_cupy_metainfo(DMatrixT):
 @pytest.mark.skipif(**tm.no_sklearn())
 def test_cupy_training_with_sklearn():
     np.random.seed(1)
-    cp.random.seed(1)
+    cp.random.seed(np.uint64(1))
     X = cp.random.randn(50, 10, dtype="float32")
     y = (cp.random.randn(50, dtype="float32") > 0).astype("int8")
     weights = np.random.random(50) + 1
@@ -210,7 +210,7 @@ class TestFromCupy:
 
     @pytest.mark.skipif(**tm.no_cupy())
     def test_qid(self):
-        rng = cp.random.RandomState(1994)
+        rng = cp.random.RandomState(np.uint64(1994))
         rows = 100
         cols = 10
         X, y = rng.randn(rows, cols), rng.randn(rows)

tests/python-gpu/test_gpu_prediction.py

@@ -226,7 +226,7 @@ class TestGPUPredict:
         cols = 10
         missing = 11  # set to integer for testing
 
-        cp_rng = cp.random.RandomState(1994)
+        cp_rng = cp.random.RandomState(np.uint64(1994))
         cp.random.set_random_state(cp_rng)
 
         X = cp.random.randn(rows, cols)
@@ -546,7 +546,7 @@ class TestGPUPredict:
 
         rows = 1000
         cols = 10
-        rng = cp.random.RandomState(1994)
+        rng = cp.random.RandomState(np.uint64(1994))
         orig = rng.randint(low=0, high=127, size=rows * cols).reshape(rows, cols)
         y = rng.randint(low=0, high=127, size=rows)
         dtrain = xgb.DMatrix(orig, label=y)
@@ -576,10 +576,10 @@ class TestGPUPredict:
         # boolean
         orig = cp.random.binomial(1, 0.5, size=rows * cols).reshape(rows, cols)
         predt_orig = booster.inplace_predict(orig)
-        for dtype in [cp.bool8, cp.bool_]:
-            X = cp.array(orig, dtype=dtype)
-            predt = booster.inplace_predict(X)
-            cp.testing.assert_allclose(predt, predt_orig)
+
+        X = cp.array(orig, dtype=cp.bool_)
+        predt = booster.inplace_predict(X)
+        cp.testing.assert_allclose(predt, predt_orig)
 
         # unsupported types
         for dtype in [