add lut_cost() function to SparseLinearNeq

[oliviaweng]

Approximate how many 6:1 LUTs are needed to implement this layer using LUTCost() as defined in LogicNets paper FPL'20:

• $LUTCost(X, Y) = (Y / 3) * (2^{(X - 4)} - (-1)^X)$

where:

• $X$: input fanin bits
• $Y$: output bits

[nickfraser]

Could you move the get_lut_cost helper function to util.py?

[nickfraser]

Looks good - thanks!