Framework-vs-framework token benchmarks.

Each comparison fixes a task suite, holds the model and tooling constant, and lists every verifiable submission received against it. New comparisons are added as task suites are drafted and reviewed (see the methodology for how a pairing qualifies).