post distributed verification

[dshulyak]

the intuition is instead of verifying all post indexes node can verify a small subset of them and rely on other nodes in the network to verify the rest.

we already know syncing atxs for several epochs takes non-negligible amount of time (#5182), after a year it may become an obstacle to joining network.

distributed verification will significantly (10x - 30x) reduce cpu requirements, technical implementation required a few changes:

• pick a random subset of indexes, randomness is seeded locally, for example by your id
• node has to prepare special-purpose malfeasence proof when invalid index is found , with that index
• for initial atx we verify all indexes, and reject atx if it doesn't pass

it relies on correct behavior for syncing late objects that is not implemented and should be a part of spacemeshos/pm#272 work.

TODO:

• Allow seeding k3 randomness and return index of invalid label post-rs#165
• verify selected indices of POST and create malfeasance proofs #5376

[dshulyak]

it relies on improvements tracked in sync v2. if we are not validating all indexes ourselves that means that we can miss invalid atx. it will skew consensus weight calculation. to address it we want to have a guarantee that if peer has a malfeasence proof then our node will receive that malfeasence proof eventually as well, with short delay that depends on implementation. there is no such guarantee with existing sync implementation.

related items: #5306