[BUG] Nondeterministic behavior in groupby-idxmax/idxmin operations

**Describe the bug**
I'm not sure if this is a true bug, but I _sometimes_ get different results after doing a groupby-idxmax operation. I'd appreciate it if someone else can reproduce too.

**Steps/Code to reproduce bug**
I haven't narrowed the bug down any further, but I've been able to reproduce the bug when `n=200` in the snippet below.
```python
import cudf
from cudf.testing import assert_frame_equal
df = cudf.DataFrame({
"a": [4, 0, 2, 2, 1, 4, 4, 4, 2, 1, 1, 1, 3, 3, 4, 1, 4, 4, 1, 0, 1, 2, 3, 4, 4, 4, 1, 2, 3, 2, 4, 2, 2, 2, 2, 2, 2, 3, 2, 1, 2, 2, 0, 2, 0, 3, 3, 0, 1, 2],
"b": [1, 1, 4, 0, 3, 3, 0, 0, 1, 3, 2, 0, 2, 3, 2, 3, 4, 3, 1, 0, 1, 2, 4, 3, 1, 4, 2, 3, 0, 0, 1, 4, 3, 4, 3, 2, 2, 2, 4, 2, 4, 3, 4, 2, 3, 1, 2, 1, 0, 0]
})

n=200
for _ in range(n):
df1 = df.groupby("a").idxmax()
df2 = df.groupby("a").idxmax()
assert_frame_equal(df1, df2) # true for most iterations (but sometimes fails)
```
```
AssertionError: Column name="b" are different

values are different (100.0 %)
[left]: {'[np.int64(42), np.int64(33), np.int64(16), np.int64(4), np.int64(22)]'}
[right]: {'[np.int64(42), np.int64(2), np.int64(16), np.int64(4), np.int64(22)]'}
```

**Expected behavior**
I expect the dataframes to be the same.

**Additional context**
I found this by collecting fallback logs in the groupby tests in the pandas test suite.