Benchmarks for attributing committer performance

[liran-funaro]

Three of the findings in this umbrella were invisible until the corresponding benchmark existed, and
one of them stopped a change that would have been made for nothing. A cluster tells you the pipeline
got slower; it does not tell you which stage, and a nineteen-machine baseline moves about 15% between
days, so small effects are only measurable in-process.

This issue tracks the set; each benchmark is a child, independently reviewable.

• [sidecar] Benchmark the whole service end to end #792 — end-to-end sidecar
• [coordinator] Benchmark the signature verifier manager #793 — coordinator signature verifier manager
• [loadgen] Benchmark the submit path #794 — load generator submit path
• [loadgen] Sweep transaction generation over core count #795 — load generator generation sweep
• [coordinator] Sweep the dependency graph benchmark over the constructor pool #796 — repair the dependency graph benchmark

Shared prerequisite: several test helpers must widen from *testing.T to testing.TB so benchmarks
can reuse them (utils/test.WaitForConnections, mock.StartMockVerifierService, and the coordinator
and sidecar test environments). Precedent exists in mock/test_exports.go, where
StartMockCoordinatorService and NewOrdererTestEnv are already testing.TB.