fix(cost-model): add true above-diagonal samples to valueContains benchmark

[Unisay]

Summary

Fixes #7525

The valueContains cost model uses const_above_diagonal which requires benchmark data in two regions:

• Above diagonal (y > x): contained size > container size → constant cost (early exit)
• Below diagonal (x >= y): container size >= contained size → linear cost

However, the benchmark function only generated samples in the below-diagonal region (contained was always a subset of container).

Changes

Benchmark Generation

• Removed misleading aboveDiagonalCombos variable (referred to Value structure diagonal, not costing diagonal)
• Added 20 true above-diagonal samples where contained.totalSize > container.totalSize with varied size ratios:
  • Small containers (4-16 entries) with larger contained values
  • Medium containers (25-100 entries) with larger contained values
  • Large containers (150-1024 entries) with much larger contained values
  • Very large containers (800-1024 entries) with extreme contained values
• Updated comments to clarify two different diagonal concepts:
  • Value structure diagonal: numPolicies vs tokensPerPolicy (internal map shape)
  • Costing diagonal: container totalSize vs contained totalSize (used by cost model)

Cost Model Updates

• Re-ran remote costing benchmarks with expanded above-diagonal coverage (20 samples)
• Updated builtinCostModelA/B/C.json with refined parameters:
  • Above-diagonal constant: 233391 → 246085 (+5.4%)
  • Below-diagonal intercept: 658617 → 686640 (+4.3%)
  • Slope parameters adjusted for better fit
• Updated benching-conway.csv with new benchmark data (~570 total points)

[kwxm]

Let's have a few more points above the diagonal!

[kwxm]

Maybe 10 or 20 with some of them containing larger values would be better, just to make sure that we get a decent average: 3 points seems a bit small.

[Unisay]

Done.