R4R F1 mechanism rounding fix

[rigelrozanski]

REF #3750

due to order of operations F1 was actually calculating stake slightly differently than staking... see the diff in the code comments for more explanation

fixes problem found here:
#3750 (comment)

Also added a defensive minimum if rewards > outstanding (which could occur in the old code if everyone withdrew in the same block)

(feel free to push to this branch)

[rigelrozanski]

unforch the a test case now fails due to the correction factor - so we should patch it however that's a bit ugly - a better solution here may be to fundamentally modify how F1 works so this form of rounding error doesn't show up... at least we know what's causing it.

feel free to push to dis branch @cwgoes

[cwgoes]

A few questions, also this doesn't seem to fix the simulator.

[cwgoes]

Simulations are failing due to division-by-zero on line 72 of x/stake/keeper/delegation.go:

  roundingCorrectionFactor := currentStake.QuoTruncate(currentStakeF1)

[cwgoes]

FYI @rigelrozanski - once the division by zero is removed, this fixes the simulator except for import/export - #3798 (comment).

I prefer the approach in #3788 (comment) though - and I even prefer just the MinSet safety check (which also fixes the sim) over applying this correction on every slash event, because I think applying this correction every time could lead to substantial reward underestimates.

[rigelrozanski]

@cwgoes we're in agreement I will implement now

[codecov]

Codecov Report

Merging #3788 into cwgoes/outstanding-per-validator will decrease coverage by 0.03%.
The diff coverage is 37.25%.

@@                         Coverage Diff                         @@
##           cwgoes/outstanding-per-validator   #3788      +/-   ##
===================================================================
- Coverage                             61.33%   61.3%   -0.04%     
===================================================================
  Files                                   189     189              
  Lines                                 14156   14167      +11     
===================================================================
+ Hits                                   8683    8685       +2     
- Misses                                 4913    4922       +9     
  Partials                                560     560

[cwgoes]

Basic logic looks good; a few minor comments & the import/export sim still fails.

[alexanderbez]

Changes look reasonable to me and @cwgoes covers anything I'd have to add :)

[rigelrozanski]

Alright so I finally got to the bottom of the bug seen in test_sim_gaia_simulation_after_import - NOTE that this bug was not introduced in this PR but in the parent PR cwgoes/outstanding-per-validator ( #3750)

see the solution in commit d9e5211

Basically what is happening is because we switched to storing outstanding per validator instead of per the whole network, when we initialized for zero-genesis-export we were reinitializing (resetting to zero) the distribution-validator record which held the scrap "decimal" amounts of outstanding coins. Normally these scraps are kept in the validator to be withdrawn later once they reach a full coin amount.

My solution just has these scraps donated to the community pool during zero-height export.

Changes addressed.