Native implementation for ARM (Neon) and x64 (AVX2, AVX-512) of HNSW Lucene bulk scoring

• Use new bulk scoring dot product for max inner product #139409
• [SIMD][ARM] Optimized native bulk dot product scoring for Int7 #138552
• [SIMD][x64] Optimized native bulk dot product scoring for Int7 #139069
• Define bulk operations for vector square distance #139198
  • TODO: sqr optimized (e.g. prefetching) -- common function/template for optimized bulk operations?
• [ ] TODO: cos EDIT: cos is not needed