Turn near-duplicate score test into a property-based test

After updating the near-duplicate scores, a test was added to ensure that near-duplicate examples have worse scores than non-near-duplicates. RIght now, the test only works on a small, toy dataset. Turning it into a property-based test would be ideal.

So that the test doesn't take too long, it may be a good idea to make a different embedding strategy that:

1. Generates a random array
2. Selects a handful of random examples to have near-duplicates. Append a new example that takes the original example with variable noise.

   # append to existing feature array
   x[sample_ids] + small_random_noise
   

3. The embedding strategy should always generate "proper" near-duplicates that will get flagged. We don't want to test cases where there are no near-duplicates.

Below is the existing test case for comparison. We want to make sure that the issue scores created by the issue manager don't violate this property.

https://github.com/cleanlab/cleanlab/blob/23af72a2bd260be3b39cf35f66a8c187ebc4d1d2/tests/datalab/issue_manager/test_duplicate.py#L89-L98

Originally posted by @elisno in https://github.com/cleanlab/cleanlab/pull/943#discussion_r1442401494