[EXAMPLE] Hybrid search with cuVS + cuDF

Hybried search is an important capability that arises in end-to-end vector databases. We have been asked now by several users whether or not metadata can be stored alongside the vectors and the usual answer is "it can, but you'd want to use soemthing like cuDF/Pandas for that". Databases like LanceDB do soemthing similar, for example, and similar to cuDF, they offer filtering / reductions / groupbys but not a first-class SQL dialect.

cuVS allows for pre-filtered search, which is often a requirement for hybrid search. Post-filtered search can also be preferred under certain circumstances, though this is often something users do themselves downstream.

cuDF offers a really powerful metadata management API and we should explore using it to implement hybrid search capabilities. At the very least, we can provide an example notebook and maybe compare cuVS+cuDF side-by-side with HNSWlib+Pandas for a direct GPU/CPU comparison. If this capability proves useful enough, we could even implement a first-class API within cuVS itself to perform hybrid search.

Start with the proposed example notebook comparing cuVS with cuDF against HNSWlib with Pandas; the issue does not name an existing file or test. Establish whether the deliverable is only the comparison or a first-class cuVS API, then document hybrid filtering and metadata handling in the chosen example or implementation.