[FEA] Adding attributes to vectors for indexing and retrieval.

Is your feature request related to a problem? Please describe.

We want to be able to supply some search conditions on predefined attributes on vectors.
We need to retrieve nearest N vectors (and attributes) after applying these search conditions.

Describe the solution you'd like

When creating the vector index, allow user to supply a small number (8 to 10 will be enough) of attributes of each vector. Each attributes is an 64 bit integer. Later during query for N nearest vectors, allow user to pass in a search condition on these attributes. Easiest conditions could be a few AND-ed comparisons, such as
attribute[0] = XXX AND attribute[3] > YYY. Include all attributes in search result as well.

Describe alternatives you've considered

More general search conditions can be useful, such as allow user to supply a UDF, taking attributes as parameter.

Additional context

MatrixOrigin/MatrixOne is a general purpose database with native support for vector data type. Our query workload is often "mixed", that is, some boolean search condition and retrieve top/nearest N vectors that satisfies these boolean search. Current cuVS allow us to supply a bitmap of vector ids but computing this bitmap is not necessarily cheap -- it forces our query engine to process search conditions first, then asking top N from vector index. Pushing these search conditions down into cuVS is essential for query performance.

The issue names no files, tests, or entry points. Start by locating the vector-index creation and nearest-N query APIs, then compare their existing vector-ID bitmap filtering path. Done should include a scoped design for 8–10 int64 attributes, AND-ed comparisons, returning attributes with results, and tests covering filtering and retrieval.