test(cuda.core): share one DummyUnifiedMemoryResource per batched test

Both batched-advise tests previously created a throwaway
DummyUnifiedMemoryResource per allocation inside a list comprehension:

    bufs = [DummyUnifiedMemoryResource(device).allocate(size) for _ in range(2)]

The Buffer holds a reference to the throwaway MR via mr=self, so the MR
should stay alive — but on CUDA 12.9.1 CI test_batched_same_advice fails
with bufs[0] showing ptr=0x0 size=0 (the post-close state). On CUDA 13
the same pattern works.

Switch to one MR shared across both allocations. This is cleaner anyway
and removes the throwaway-per-iteration pattern as a possible source of
the cu12 issue. If the failure persists, we'll know the MR lifetime
wasn't the cause.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: rparolin

cuda_core/tests/test_memory.py

@@ -2242,7 +2242,8 @@ def test_batched_same_advice(self, init_cuda):
         device = Device()
         _skip_if_managed_location_ops_unsupported(device)
         device.set_current()
-        bufs = [DummyUnifiedMemoryResource(device).allocate(_MANAGED_TEST_ALLOCATION_SIZE) for _ in range(2)]
+        mr = DummyUnifiedMemoryResource(device)
+        bufs = [mr.allocate(_MANAGED_TEST_ALLOCATION_SIZE) for _ in range(2)]
         advise(bufs, "set_read_mostly")
         for buf in bufs:
             assert (
@@ -2260,7 +2261,8 @@ def test_batched_per_buffer_location(self, init_cuda):
         device = Device()
         _skip_if_managed_location_ops_unsupported(device)
         device.set_current()
-        bufs = [DummyUnifiedMemoryResource(device).allocate(_MANAGED_TEST_ALLOCATION_SIZE) for _ in range(2)]
+        mr = DummyUnifiedMemoryResource(device)
+        bufs = [mr.allocate(_MANAGED_TEST_ALLOCATION_SIZE) for _ in range(2)]
         advise(
             bufs,
             "set_preferred_location",