Making multiplication and inner product more efficient

[goforashutosh]

Hi, we used this repo for matrix multiplication, which involved a large number of multiplications and inner products. The diff below shows a lot of changes due to the formatting differences in our code, but this fork only modifies two functions, which were in the original repo- signed_div_scale and inner_product.

For signed_div_scale, we implement |q|<2^3p bound by translating q by 2^3p and doing a range check. This reduces the number of cells required per multiplication call from ~250 to ~90 (for lookup bits = 12).

For inner_product, we change the implementation so that signed_div_scale is called only once since this is the expensive call. This makes the number of cells go from 90*N to N+90 where is the size of the vectors for inner product.