Delay power until first window PoSt

[Stebalien]

This will need a lot of tests, and the existing tests are now all broken.

Design

Every partition gets a new Unproven bitfield for sectors that have yet to be proven and an UnprovenPower field store power to be proven and activated when the partition is next proven.

Crucially, new faulty power from unproven sectors is never removed from the miner, because it was never activated. Furthermore, unproven power is added to the miner when it's proved the first time. To achieve this, many methods have gained poweDelta return values.

Faults

When marked faulty, unproven sectors are removed from the unproven bitfield and added to the faulty bitfield. They now join the normal fault flow.

Termination

Unproven sectors may be terminated (early) like any other sector.

WindowPoSt

On WindowPoSt, skipped unproven sectors are added to the faulty sectors bitfield/power normally. Then, UnprovenPower and the Unproven bitfield are zeroed.

End of deadline

At the end of the deadline, unproven sectors in unposted partitions are marked faulty along with their power.

Handling sector expiration before handling unproven sectors is explicitly forbidden.

[anorth]

Epic, looking good. FYI @nicola

[Stebalien]

This still needs more tests, but it's now passing the existing tests and is ready to review.

[codecov-commenter]

Codecov Report

Merging #883 into master will decrease coverage by 0.0%.
The diff coverage is 62.0%.

@@           Coverage Diff            @@
##           master    #883     +/-   ##
========================================
- Coverage    74.0%   74.0%   -0.1%     
========================================
  Files          57      57             
  Lines        6483    6550     +67     
========================================
+ Hits         4803    4847     +44     
- Misses       1083    1092      +9     
- Partials      597     611     +14     

[Stebalien]

Ok, I've found a bug. I'm not correctly handling power when handling missing posts at the end of the proving deadline. I'll upload a fix with tests tomorrow.

[Stebalien]

Unrelated change to the testing framework to make reading errors easier.

[Stebalien]

@anorth the second commit fixes the bug (and adds/fixes more tests). I'm now explicitly returning:

• PowerDelta (not all newely faulty power is part of this delta, as some of it may not have been "active").
• PenaltyPower (retracted power + newly faulty power)
• NewFaultyPower

[anorth]

Thanks.

It's slightly unfortunate that the "policy" logic of which power to penalize is now in the state, but not a huge problem. We could probably resolve this and make things a bit more concise with a "power deltas" struct for use as return value from the methods that return more than one power thingy.

[Stebalien]

Hm. Yeah, adding a power delta object would clean things up a bit (#1011).