current features

GPUMatrix class contains the following working features :
- really high speed matrix*matrix multiplication (cublas version)
- high speed common binary and unary operators
- relatively high speed LU and Cholesky factorizations (nearly 7 times speed up compared to matlab)

Works in progress :
- relatively high speed QR factorizations (not fully working yet)

Work for near future :
- Arnoldi/Lanczos algos for Hessenberg transformation, and eigen values decomposition.
- linear system solving and matrix inverse