pow-migration: Fix detection of switching validator in pre-stake

[jsdanielh]

Fix the detection of switching validator in pre-stake as it was filtering always transactions that didn't have the minimum stake but for switching to another validator in pre-stake, the stake is already above the minimum since it was already counted to another validator.

This fixes #2399.

Pull request checklist

• All tests pass. The project builds and runs.
• I have resolved any merge conflicts.
• I have resolved all clippy and rustfmt warnings.

[Eligioo]

@jsdanielh I do think this is still prone to incorrect outcomes. I believe decrease_validator_total_stake should only be set to Some if the staker is actually switching between validators. It can very well be that the staker doesn't change validator and only simply increase his stake. In this scenario the stakers.entry will be Occupied as you process the top-up transaction after his initial pre-stake transaction. But right now this will result in validator.total_stake to reduced in the end, which shouldn't happen.

To make this easier to reason about, and since stake can jump around validators quite a lot, is that you compute the total stake of a validator in a second loop after you've done processing all the transactions in for (_, txns) in txns_by_sender.iter_mut(). Instead of trying to keep track of the total stake as you process the txns_by_sender.

[jsdanielh]

@jsdanielh I do think this is still prone to incorrect outcomes. I believe decrease_validator_total_stake should only be set to Some if the staker is actually switching between validators. It can very well be that the staker doesn't change validator and only simply increase his stake. In this scenario the stakers.entry will be Occupied as you process the top-up transaction after his initial pre-stake transaction. But right now this will result in validator.total_stake to reduced in the end, which shouldn't happen.

To make this easier to reason about, and since stake can jump around validators quite a lot, is that you compute the total stake of a validator in a second loop after you've done processing all the transactions in for (_, txns) in txns_by_sender.iter_mut(). Instead of trying to keep track of the total stake as you process the txns_by_sender.

Makes sense. Just made the change to implement this.