solana-program: improve VoteState::deserialize_into()

[2501babe]

Problem

earlier this year we implemented VoteState::deserialize_into(), a custom parser intended to be suitable for usage in a bpf context. we want to use the new parser everywhere because, with some optimizations, it is 7-20x faster than bincode depending on the input

HANA custom vs bincode
* we win: 10000, they win: 0
* our max 22.422µs, their max 471.89µs
* our avg 5.76µs, their avg 35.956µs
* avg speedup 30.195µs

it also uses half as much memory for V1_14_11 because conversion to Current happens during parsing

Summary of Changes

implement those optimizations, and also remove some constraints that make this parser behave differently from bincode::deserialize. this allows VoteState::deserialize_into to be used in optimized vote processing

in the future we would like to replace VoteState::deserialize to use deserialize_into for everything, but this requires testing on mainnet-beta because it affects sensitive parts of the codebase which cannot be feature-gated

part of solana-labs#35101

[alessandrod]

Looks great! A couple of nits and see the CircBuf comment

[alessandrod]

if is_empty=true and prior_voters.buf is filled, we diverge from bincode
i think? One could maliciously create such a vote state.

(I think our Arbitrary impl for CircBuf should probably additionally generate this)

[2501babe]

its true that it would diverge. how could one create such a vote state tho? i wrote this based on the fact that it seemed there was no code path to do that, all changes to circbuf go through the insert function (hence also why the Arbitrary impl works that way)

[alessandrod]

I’m a hacker, I see this commit get deployed to mainnet, as soon as it is I craft a special payload to split the cluster

[2501babe]

i was gonna say some stuff about how such a payload cant be delivered because it would fail vote program ownership, but it turns out theres negligible perf benefit to the special logic anyway. so, onwards!

[alessandrod]

In the call site where I'm looking to use this (Bank::update_stakes_cache), I'm definitely seeing votes fail to deserialize, so surely this must be possible somehow?

[2501babe]

not sure what itd be withput seeing an example. in any event, should be the same as bincode now

[alessandrod]

great work!