fix(miner)!: remove DEAL_WEIGHT_MULTIPLIER and its input to QAP calc

[rvagg]

Ref: #1573

As outlined in #1573, as long as QUALITY_BASE_MULTIPLIER == DEAL_WEIGHT_MULTIPLIER then we factor out deal_space in QAP calculations. It appears to me from the language of the spec that it was imagined that the "DealWeightMultiplier (DWM)" was something that could be tuned but unless anyone imagines a world where we tinker with anything other than the "VerifiedDealWeightMultiplier (VDWM)" then we should just do away with deal_weight entirely in the calculations to avoid confusion and overhead.

In terms of what this does to quality_for_weight, here's what I see in that code as it is currently. Where WSS is weighted_sum_space_time, S is size, D is duration, W is deal_weight and V is verified_deal_weight, we end up doing this:

WSS = 10 * (S * D - W - V) + 10 * W + 100 * V
WSS = 10 * S * D - 10 * W - 10 * V + 10 * W + 100 * V
WSS = 10 * S * D - 10 * V + 100 * V
WSS = 10 * S * D + 90 * V

Which leads to our QAP result (without precision adjusting):

QAP = (S * D + 9 * V) / (S * D)

i.e. W appears nowhere in the simplified version. We just do lots of unnecessary calculations with it.

You can even see this in the test for the QAP calculation quality_is_independent_of_size_and_duration:

    // Quality of space with no deals. This doesn't depend on either the sector size or duration.
    let empty_quality = BigInt::from(1 << SECTOR_QUALITY_PRECISION);
    // Quality space filled with non-verified deals.
    let deal_quality = &empty_quality * (DEAL_WEIGHT_MULTIPLIER.clone() / QUALITY_BASE_MULTIPLIER.clone());

Substitute * (DEAL_WEIGHT_MULTIPLIER.clone() / QUALITY_BASE_MULTIPLIER.clone()) with * (10/10).

It would be nice to take this simplification further and remove all "unverified space" calculations, but SectorOnChainInfo still needs it and I guess we should keep it accurate unless we FIP to remove it or set it to zero.

[Stebalien]

This looks correct but... it would be nice to run a test on mainnet where we verify that it doesn't change power calculations in any way (if that's not too horribly difficult...).

[Stebalien]

My concern is mostly due to potential rounding/scaling issues...

[anorth]

Great, thank you.

I think it would be prudent to add a few tests that spell out the old calculation and verify it gets exactly the same results as the newly modified function.

[rvagg]

@anorth I added back in original_quality_for_weight which retains the original form (minus the constant) and pulled it in to quality_is_independent_of_size_and_duration to verify that it gets the same result.

[rvagg]

Updated with even more tests against the original form. Also matching the same tests in filecoin-project/go-state-types#308 but over there we get to test against the actual v14 form of it, which is nice.