[BUG] CagraIndex with IVF-PQ fails with too many invalid or duplicated neighbor nodes

Describe the bug
When building a CagraIndex with IVF_PQ algorithm on a dataset of 768 dims, I am getting a failure " too many invalid or duplicated neighbor nodes". If I change the algorithm to "NN_DESCENT", the building of the graph succeeds.

Using Java API with default params for IVF.

  CagraIndexParams params = new CagraIndexParams.Builder().withNumWriterThreads(1) 
            .withCagraGraphBuildAlgo(CagraIndexParams.CagraGraphBuildAlgo.IVF_PQ)
            .withGraphDegree(M)
            .withIntermediateGraphDegree(intDegree)
            .withMetric(distanceType)
            .build();

 var indexBuilder = CagraIndex.newBuilder(cuVSResources).withDataset(dataset).withIndexParams(params);
var index = indexBuilder.build();

Caused by: java.lang.RuntimeException: cuvsCagraBuild returned 0[RAFT failure at file=/home/mayya/Elastic/cuvs/cpp/src/neighbors/detail/cagra/graph_core.cuh line=1417: Could not generate an intermediate CAGRA graph because the initial kNN graph contains too many invalid or duplicated neighbor nodes. This error can occur, for example, if too many overflows occur during the norm computation between the dataset vectors.
Obtained 6 stack frames
#1 in /home/mayya/Elastic/cuvs/cpp/build/install/lib/libcuvs_c.so: raft::logic_error::logic_error(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) +0x94 [0x731229877944]
#2 in /home/mayya/Elastic/cuvs/cpp/build/install/lib/libcuvs.so: void cuvs::neighbors::cagra::detail::graph::optimize<unsigned int, raft::host_device_accessor<std::experimental::default_accessor<unsigned int>, (raft::memory_type)0> >(raft::resources const&, std::experimental::mdspan<unsigned int, std::experimental::extents<long, 18446744073709551615ul, 18446744073709551615ul>, std::experimental::layout_right, raft::host_device_accessor<std::experimental::default_accessor<unsigned int>, (raft::memory_type)0> >, std::experimental::mdspan<unsigned int, std::experimental::extents<long, 18446744073709551615ul, 18446744073709551615ul>, std::experimental::layout_right, raft::host_device_accessor<std::experimental::default_accessor<unsigned int>, (raft::memory_type)0> >, bool, bool) +0x1ed1 [0x731176710701]
#3 in /home/mayya/Elastic/cuvs/cpp/build/install/lib/libcuvs.so: cuvs::neighbors::cagra::index<float, unsigned int> cuvs::neighbors::cagra::detail::build<float, unsigned int, raft::host_device_accessor<std::experimental::default_accessor<float const>, (raft::memory_type)0> >(raft::resources const&, cuvs::neighbors::cagra::index_params const&, std::experimental::mdspan<float const, std::experimental::extents<long, 18446744073709551615ul, 18446744073709551615ul>, std::experimental::layout_right, raft::host_device_accessor<std::experimental::default_accessor<float const>, (raft::memory_type)0> >) +0x93d [0x73117672c5ed]
#4 in /home/mayya/Elastic/cuvs/cpp/build/install/lib/libcuvs.so: cuvs::neighbors::cagra::build(raft::resources const&, cuvs::neighbors::cagra::index_params const&, std::experimental::mdspan<float const, std::experimental::extents<long, 18446744073709551615ul, 18446744073709551615ul>, std::experimental::layout_right, raft::host_device_accessor<std::experimental::default_accessor<float const>, (raft::memory_type)0> >) +0x35 [0x7311766f3205]
#5 in /home/mayya/Elastic/cuvs/cpp/build/install/lib/libcuvs_c.so: cuvsCagraBuild +0x410 [0x7312298a4280]
#6 in [0x731357e2a227]
]
        at com.nvidia.cuvs.internal.common.Util.checkCuVSError(Util.java:74) ~[?:?]
        at com.nvidia.cuvs.internal.CagraIndexImpl.build(CagraIndexImpl.java:194) ~[?:?]
        at com.nvidia.cuvs.internal.CagraIndexImpl.<init>(CagraIndexImpl.java:81) ~[?:?]
        at com.nvidia.cuvs.internal.CagraIndexImpl$Builder.build(CagraIndexImpl.java:777) ~[?:?]
        at com.nvidia.cuvs.internal.CagraIndexImpl$Builder.build(CagraIndexImpl.java:721) ~[?:?]

Start with the CAGRA graph-building failure at cpp/src/neighbors/detail/cagra/graph_core.cuh:1417 and trace how IVF_PQ differs from NN_DESCENT. Review the Java path in CagraIndexImpl.java, especially the build call around lines 194 and 721-777. Reproduce the 768-dimensional dataset case and identify a reliable fix or documented handling for the invalid or duplicated neighbor error.