Update mempool 1559 params

[ValarDragon]

This PR updates mempool 1559 params to update a little slower. Its reasoned about, due to low gas usage txs on mainnet being observed to be around 30M gas, so were not lowering fast enough, and were spiking too fast.

As a comparable ETH block time (15s) is 2.5x slower than us, and they update at a maximum +- of 12.5%. Under spam we are currently updating at (1 + (50/70)/10)**2.5 = 18.8% upwards. We currently update at 0% usage for 15 seconds at -23%. However we do not see 0% usage, we instead see usage around 20-40M, so lets say 30M. That puts us at (1 + (-40/70)/10)**2.5 = -13.7% .

So we see this asymmetrical upwards increase in our real settings. Lets assume that 30M is userflow/relayers, who are getting inputted Keplr's much higher gas numbers.

I suggest changing this target block gas param to 75M, which would then change

• max update per block to: 6%
• max update over 15 seconds to: 15.6%
• max decrease per block (unchanged)
• expected decrease per block (30M): -14.4%

However, the spam / old txs are at a lower fee. They only gets evicted once base fee exceeds recheck_factor * spam_fee_rate. Currently this is set at 4 for high conservatism, especially around clients who haven't patched. Almost all clients in the ecosystem have patched, so we can reduce this. Also clients have standardized around using a fee much higher than base fee.

This PR changes it 3. 3.3 gives the same guarantee to clients under max load with the above changes. (Follow the log line in the doc).

The change to 3 will make max gas usage take 19 blocks to evict instead of 21. I believe we can get away with far lower to be honest, due to wallets over-estimating fee right now.

The effects of this are:

• I believe the recheck factor is what controls the height of the peaks. This reduction from 4->3, should cause a 25% reduction I hope, and probably better.
• Lowering the max rate of change will smoothen out the curves, and lower "edge effects" causing higher tops than we need.
• Raising target gas will help improve recovery time once we've past the over-filling point.
• We may see periods of high block gas due to spam take a few more seconds to clear out

[ValarDragon]

Osmosis mainnet basefee for the last 3 hours (ignore the blue line)

You see the high spikyness, and the downside being slower to come down than upside.

[ValarDragon]

This was just overdue. Have gotten multiple points of feedback from integrators asking for this. It doesn't really matter either way. Highest we've seen is .07.

[nicolaslara]

This generally makes sense to me, but I think it's one of those things where we need to test it and see how it behaves under real load

[nicolaslara]

code looks good