- Add a new micro benchmark for reduced precision type Int8 and compute the dot-product at 32-bit accumulation.
- Compare its performance against byte array-backed backing storage.
- Exiting flat array allocation routines has a native implementation.

There is a considerable performance gap w.r.t primitive array-based backing storage.

Benchmark (size) Mode Cnt Score Error Units
Int8FlatArrayDotProduct.int8DotProductWith32BitAccumArray 1024 thrpt 2 8237.625 ops/s
Int8FlatArrayDotProduct.int8DotProductWith32BitAccumArray 2048 thrpt 2 4272.565 ops/s
Int8FlatArrayDotProduct.int8DotProductWith32BitAccumFlatArray 1024 thrpt 2 57.959 ops/s
Int8FlatArrayDotProduct.int8DotProductWith32BitAccumFlatArray 2048 thrpt 2 29.011 ops/s