test(bench): prove parity and retire legacy scale orchestration

[DecisionNerd]

Objective

Compare the completed ReFrame/BenchExec/Fly ladder with preserved legacy orchestration and retire only duplicated behavior after semantic parity is demonstrated.

Requirements

• Preserve exact legacy outcomes as migration fixtures.
• Use completed test(scale): complete the Fly S18-S26 lifecycle ladder #900 ladder evidence for provider semantics; do not authorize a separate duplicate scale run solely for parity.
• Compare phases, correctness, failure classes, RSS, storage, logical/physical I/O, reader/publication counters, timing, provenance, and teardown.
• Run no-cost tiny/local shadow fixtures first.
• Explain every semantic difference before cutover.
• Retain bounded correctness tests in product CI.

Acceptance criteria

• Parity matrix has no unexplained semantic gaps.
• New harness becomes authoritative only after accepted tiny-fixture and completed-ladder comparison.
• Removed orchestration has equivalent external coverage.
• Historical evidence remains readable.
• No additional S18-S26 provider ladder is run for this issue.

BDD completion scenario

Given identical tiny fixtures, preserved legacy results, and completed #900 evidence
When evidence is normalized and compared
Then semantics agree or every difference is accepted
And duplicated legacy orchestration is retired without losing coverage.

Relationships

• Parent: epic(bench): externalize scale certification and add GDC benchmark suites #952
• Requires profiles/controller test(bench): define the complete S18-S26 qualification ladder #956, Fly adapter build(bench): provide the disposable Fly ladder adapter #958, and completed ladder test(scale): complete the Fly S18-S26 lifecycle ladder #900.