sealing: Check piece CIDs after AddPiece

[magik6k]

Non-worker-api-breaking alternative to #7175

The difference here is that after adding a bad piece ta a sector, it will be market as 'allocated' in the sector file, but that's actually fine as we allow overwriting allocated parts of sectors - the only thing we need to do to recover from this is to not add the bad piece to the sector.Pieces array, and the subsequent AddPiece calls will overwrite the bad data with good data

[codecov]

Codecov Report

Merging #7185 (b42171d) into master (c664f9e) will increase coverage by 5.22%.
The diff coverage is 8.33%.

@@            Coverage Diff             @@
##           master    #7185      +/-   ##
==========================================
+ Coverage   33.80%   39.02%   +5.22%     
==========================================
  Files         601      607       +6     
  Lines       64277    64577     +300     
==========================================
+ Hits        21728    25204    +3476     
+ Misses      38464    34985    -3479     
- Partials     4085     4388     +303     

Impacted Files	Coverage Δ
extern/storage-sealing/input.go	56.45% <0.00%> (-1.94%)	⬇️
extern/storage-sealing/states_sealing.go	37.87% <33.33%> (-0.03%)	⬇️
storage/addresses.go	55.95% <0.00%> (-4.77%)	⬇️
storage/wdpost_sched.go	77.22% <0.00%> (-3.97%)	⬇️
extern/sector-storage/sched.go	83.05% <0.00%> (-1.66%)	⬇️
node/impl/storminer.go	22.46% <0.00%> (-0.40%)	⬇️
extern/sector-storage/stores/local.go	59.25% <0.00%> (ø)
chain/actors/builtin/paych/mock/mock.go	64.28% <0.00%> (ø)
markets/dagstore/mocks/mock_lotus_accessor.go	80.48% <0.00%> (ø)
chain/vectors/gen/main.go	2.36% <0.00%> (ø)
... and 134 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c664f9e...b42171d. Read the comment docs.

[Stebalien]

Both solution seem to work fine, although the first one is arguably safer.

[Stebalien]

Coverage should hopefully be fixed by #7189.
Broken integration test is disabled in #7187.

[Stebalien]

Ok, integration tests disabled entirely.

[magik6k]

although the first one is arguably safer

It's not really safer, we only really use the info about which bits of the unsealed sector file are allocated to check if we need to Unseal the sector before reading unsealed data.
Since here we wrote bad data into the sector, we won't be unsealing that piece anyways, and it will just be overwritten with real data. The fact that this miswritten piece is marked as allocated doesn't really impact anything at the sealing stage