poly-approximation: drop very small coefficients

[j2kun]

Either in polynomial-approximation or in lower-polynomial-eval (perhaps controlled by a new pipeline-level option as in #2137), we should have the ability to drop coefficients that have very small magnitude (e.g., 1e-16). In the FHE literature these are typically dropped because they add negligible noise to the overall polynomial approximation, but keeping them introduces major scale issues as very large scaling factors are needed to represent them in FHE.