[ANN_BENCH] Better index naming strategy

While serializing indexes with all the params embedded in the filename, the filename can easily grow beyond 255 characters, leading to the indexes being saved with incorrect (machine generated random numbers) names. I noticed this for example in CAGRA when I set ivf-pq build and search parameters (so each name would have to have long substrings like ivf_pq_build_kmeans_niter10 which can add a lot of characters). We can consider coming up with a better naming convention. Perhaps keep the name simple and store an additional metadata file containing all the parameters?

Start by tracing ANN_BENCH index serialization, especially the CAGRA path where parameter names are embedded in filenames. Determine how names are generated and how serialized indexes are loaded; the work is done when a naming convention avoids the filename-length failure while preserving access to all build and search parameters.