Reduce shuffle proof serialization overhead

[dapplion]

ark-serialize https://crates.io/crates/ark-serialize is good to get some byte format going. However given that this data is spent over the wire we want minimal possible size. Since N in known ahead of time, we can concat all compressed points without any extra overhead.

My tests with https://github.com/asn-d6/curdleproofs/compare/main...dapplion:test-vectors?expand=1 show that the serialized proof size is 480 * log2(N) + 1216. Concatenating all points in the proof assuming 32 bytes per Fr, it should be 480 * log2(N) + 1136.