Cover Pruning #4
Description
Currently we evaluate all nodes of the valid prefix cover. However, this surely contains some paths that have vanishingly small probability. If our tree invariants are correct, we will always know the probability of a node before we expand its children. The next-byte distribution is built by summing the byte distributions for each node, weighted by the node probability. This means that we can take in a parameter describing the worst-case total-variation distortion allowed by pruning and prune optimally to meet that guarantee!