Add new non-indexed vector search benchmarks

[benwtrent]

Both of our main rally tracks (so-vectors and dense-vectors) fully index the vectors into Lucene's HNSW graph.

However, on < 100k scale, it is common for users to not bother indexing into the HNSW graph and simply use our scripting queries. We should measure performance improvements on indexing & search for vectors that do not get indexed into HNSW.

Additionally, this should cover both byte encoded and float encoded.