Address alignment differences with cuda::mr::resource_ref

Our implementation of `cuda::mr::{async_}resource_ref` currently aligns to `alignof(max_align_t)`

This is a minor difference to the current device_memory_resource. We intent to expand the interface of `cuda::mr::{async_}resource_ref` to take desired alignment into account.

However, this is out of scope of this PR and will be done in the cccl repository

python/rmm/tests/test_rmm.py

@@ -646,19 +646,19 @@ def test_statistics_resource_adaptor(stats_mr):
         buffers.append(rmm.DeviceBuffer(size=1000))
 
     assert mr2.allocation_counts == {
-        "current_bytes": 2000,
+        "current_bytes": 2016,
         "current_count": 2,
-        "peak_bytes": 2000,
+        "peak_bytes": 2016,
         "peak_count": 2,
-        "total_bytes": 2000,
+        "total_bytes": 2016,
         "total_count": 2,
     }
     assert stats_mr.allocation_counts == {
-        "current_bytes": 7000,
+        "current_bytes": 7056,
         "current_count": 7,
-        "peak_bytes": 10000,
+        "peak_bytes": 10080,
         "peak_count": 10,
-        "total_bytes": 12000,
+        "total_bytes": 12096,
         "total_count": 12,
     }
 
@@ -668,17 +668,17 @@ def test_statistics_resource_adaptor(stats_mr):
     assert mr2.allocation_counts == {
         "current_bytes": 0,
         "current_count": 0,
-        "peak_bytes": 2000,
+        "peak_bytes": 2016,
         "peak_count": 2,
-        "total_bytes": 2000,
+        "total_bytes": 2016,
         "total_count": 2,
     }
     assert stats_mr.allocation_counts == {
         "current_bytes": 0,
         "current_count": 0,
-        "peak_bytes": 10000,
+        "peak_bytes": 10080,
         "peak_count": 10,
-        "total_bytes": 12000,
+        "total_bytes": 12096,
         "total_count": 12,
     }
     gc.collect()
@@ -696,7 +696,7 @@ def test_tracking_resource_adaptor():
     for i in range(9, 0, -2):
         del buffers[i]
 
-    assert mr.get_allocated_bytes() == 5000
+    assert mr.get_allocated_bytes() == 5040
 
     # Push a new Tracking adaptor
     mr2 = rmm.mr.TrackingResourceAdaptor(mr, capture_stacks=True)
@@ -705,8 +705,8 @@ def test_tracking_resource_adaptor():
     for _ in range(2):
         buffers.append(rmm.DeviceBuffer(size=1000))
 
-    assert mr2.get_allocated_bytes() == 2000
-    assert mr.get_allocated_bytes() == 7000
+    assert mr2.get_allocated_bytes() == 2016
+    assert mr.get_allocated_bytes() == 7056
 
     # Ensure we get back a non-empty string for the allocations
     assert len(mr.get_outstanding_allocations_str()) > 0