poly: add a pattern to break down an NTT into lower-degree NTTs

[AlexanderViand-Intel]

Software implementations of NTT are generally pretty flexible in the degree they support, but HW implementations are more likely to have a specific size or range of sizes. Therefore, we need to add a pattern to break down NTTs at some degree 2^{k+m} into a bunch of degree 2^k NTTs.