[BUG] Unit test failure with filtered indices

[daxpryce]

For some reason the filtered index vs. the unfiltered index (in mem, not dynamic) are sporadically returning different results. the main thrust of the test is "run a query without a filter and ask for enough k neighbors that the entire set is returned (to avoid any approximate NN non-determinism), and then run it again with the filter "all" that every point got marked with, with the same k, and they should be identical sets.

Anyway, in some test runs it's just fine and every so often we only get 9,999 of the same results instead of 10k. nothing in the python code should impact this, but there may be a bug that my test is exercising that the existing cxx unit tests do not.

#592 has some commentary around this as I discovered it. It's inconsistent as to when it appears, but the recent action run on https://github.com/microsoft/DiskANN/actions/runs/11503719966/job/32021629349#step:3:2568 shows the failure. As you can see from the other python builds, it's not consistent across OS or python version, and as you can see from a later run: https://github.com/microsoft/DiskANN/actions/runs/11504921348 , all the python build steps complete successfully.

The unit test in question that exercises this scenario is: https://github.com/microsoft/DiskANN/blob/main/python/tests/test_static_memory_index.py#L270