Too many allocation failures when starting a simulation

[darosior]

With any parameter the simulation will have a lot of allocation failures at the beginning because of the (artificially) high fees around block 350k.

We should investigate why, as our refill should cover for those fees and it apparently does not. It may uncover a bug in the refill computation.

[JSwambo]

Using #75 I set up with DELEGATE_RATE=7 and UNVAULT_RATE=5 and with_fixed_scale=False to simulate an increasing capital requirement for the WT. This results in consistent delegation failures throughout the entire simulation, rather than just in the beginning, and confirms my hypothesis that delegation failures is correlated with increasing capital requirement.

A factor that significantly reduces the delegation failure rate is the REFILL_EXCESS. We can play around with this for different operating parameters to find something reasonable per-deployment.

[JSwambo]

Also, a shorter REFILL_PERIOD significantly reduces delegation failure rate.

[darosior]

Then it's probably related to a bug in how we compute the refill amount. Maybe #86 is related.