GitLab

* replacing array with tuple and vector for tensor data

updated block-cluster in gridwise gemm and thread-cluster in blockwise copy to use cluster descriptor

using tuple (instead of vector) for holding C thread matrix data to solve register over-allocation issue

* overhaul vector_type, make int8x4_t real vector instead of aliasing from int32_t

* initial implementation for magic number division and DynamicMerge_v2_magic_division that uses it

* turn off DynamicMerge_v2_magic_division that use magic number division by default