Limit the execution time of on-chain sequential phragmen

[kianenigma]

We fallback to the on-chain election if nothing else works, which is okay, but we stall the chain for a long time. This is not good. We should try and somehow make the on-chain sequential run smoother, at the cost of not including all voters.

Some thoughts on the matter:

1. we can and should sacrifice the small staked voters, but we always want to keep all candidates.
2. Currently, we don't keep a sorted list of voters anywhere, so we need to take into account that we need to sort them, and then take the top x.
3. We must first find the sweet spot where for any number of voters more than this, it is wroth it to first sort and then run phragmen.
4. We must also measure exactly how much weight seq-phragmen is taking in Benchmark exactly how much weight on-chian seq-phragmen takes.  #8347, and then we need to alter Benchmark exactly how much weight on-chian seq-phragmen takes.  #8347 or generally make it suitable, so you can query the weight before executing anything. For example, assume pallet-multi-phase-election wants to fallback to on-chain election. Before actually calling onchain::elect(...), we ideally want to know how many voters we are allowed to pass in, in order for the weight of the elect call to be below a certain threshold t. This might make the API a bit messy, but we can add new functions to trait ElectionProvider to accommodate this, namely predict_elect_weight() -> Weight where this would return a best effort guess of how much weight is to be consumed, if we call elect, with exactly the same data.

@coriolinus I wonder if you have opinions on this, and the linked issue.

[stale]

Hey, is anyone still working on this? Due to the inactivity this issue has been automatically marked as stale. It will be closed if no further activity occurs. Thank you for your contributions.

[kianenigma]

We don't run this anymore, but once we have #9081, we can configure the onchain phragmen to use only a small subset of the voters.

[kianenigma]

this is now quite possible to do with the bags list, and quote a good one to provide for the Parachains. Essentially, using staking + a new BoundedOnChainSequentialPhragmen is safe.

@emostov Fancy doing this one? I think doing a bit of documentation about it and making sure it is broadcasted to our parachain teams is also a good step.