Fix uninitialized memory issue

Call to BLAS.dscal passes wrong size, causing the resulting partition matrix to assume the size of the full matrix, with the extra data consisting of uninitialized contents. This causes sporadic errors in the matrix computation under certain circumstances.
It also can cause scale and performance problems since each C partition has the full size of the C matrix in its mat_part.
To reproduce, use a matrix with multiple partitions. Pass all zeros for the B matrix, but non-zero C matrix. That will frequently result in wrong values for the result.
To fix, replace the line:
SELF.mat_part := BLAS.dscal(map_c.matrix_rows*map_c.matrix_cols,
beta, part.mat_part, 1);
with:
SELF.mat_part := BLAS.dscal(part.part_rows*part.part_cols,
beta, part.mat_part, 1);