Add ML-KEM polynomial vector and matrix layer

Add the polynomial vector and matrix abstractions (compression,
serialization, NTT, and matrix-vector products) built on top of the
polynomial and sampling primitives.