[FEA] Vamana indexing optimizations

[cjnolet]

[bkarsin]

The optimization progress has been delayed due to bugs introduced and performance benefits not meeting expectations. Thus far I have implemented and tested many optimizations (some more details in #1757) and performance is between 5-40% faster than the baseline. In addition, I have added fp16 support that gets up to 2x speedup over fp32. I am currently working on final rounds of optimization to try and get a bit more performance before submitting the PR.