[FEATURE]Remove double converting from byte[] to float[], and float[] to byte[] for script scoring with binary vector

[heemin32]

In #1826 which support script scoring on knn binary vector, we convert byte[] to float[] and again float[] to byte[] which will add a latency during query using script scoring on knn binary vector. We want to avoid the unnecessary converting.

[navneet1v]

@heemin32 why we are not resolving this as part of 2.16 when this impacts the latency?

[heemin32]

It is not a trivial change and 2.16 code freeze is just a next Monday. Score script is already slow compared to ANN and the impact on latency due to double converting could be negligible though actual profiling might be needed to confirm.

[navneet1v]

and the impact on latency due to double converting could be negligible

till we don't have benchmarks we cannot say this. and leaving optimization out of a release doesn't seem correct to me.

[heemin32]

and the impact on latency due to double converting could be negligible

till we don't have benchmarks we cannot say this. and leaving optimization out of a release doesn't seem correct to me.

Optimization can be implemented incrementally. That is how it has been till now.