test(testkit): handle exact ambiguous outcomes in paired authority fixture

[VerifiedOrganic]

The paired authenticated-consumer fixture test failed in the required miscellaneous Rust shard on SDK PR #801 at head 1e65f822facda35d33bfa0def13d9b68f144b487.

The executed failure was authenticated_consumer_fixture::tests::fixture_pair_shares_authenticated_authority_and_reopens_without_v1_lowering at crates/opc-session-testkit/src/authenticated_consumer_fixture.rs:1903: its one opaque dispatch returned FencedTransitionExecuteError::OutcomeUnknown and the test expected immediate success. The testkit library reported 110 passed and one failed. Original CI job. Raw log SHA256: fead59df5926e265bd674d52696b66d33adb509887f92f6532eb21f49ad91108.

The fixture source is byte-identical to main at 3194cf3c06ab0f6aeed3f39b00f1055702c75f09 and the preceding two candidates. This comparison does not identify the original transport or scheduling cause. The earlier recovery failures tracked by #823 are different tests; that recovery group passed in this attempt.

The existing public contract permits exactly one mutation dispatch. A possible send removes dispatch authority and permits only read-only receipt recovery of the retained request. PreparedCheckpointBudget fixes a three-second outer budget for this test and a 250 ms physical-attempt cap. The paired fixture test proves shared authority, unchanged general-backend capabilities and journal reopening; it is not a successful-first-response latency qualification.

Acceptance:

• Preserve the original CI failure and reproduce the immediate-success assumption deterministically using the existing real post-commit response-loss fixture control.
• If the paired test receives the exact request's OutcomeUnknown, require an authoritative successful receipt through the existing status-only handle before reading shared state, within the original absolute deadline. Reject every other error and mismatched identity.
• Keep one physical mutation, zero general mutation/CAS calls, exact retained identity, successful general readback and fresh-facade status-only reopening assertions. A missing or rejected receipt must never count as success.
• Cover ordinary completion and deliberate response loss using real authenticated voters. Preserve fix-removal and a distinct adversarial control, then rerun the restored focused tests and normal required gates.
• Do not change production APIs, increase either deadline, replay or reseal the mutation, manufacture a receipt, disable coverage, or claim that a deterministic control proves the original CI timing cause.

Related contract: #696. This is a generic synthetic SDK fixture investigation and correction; no downstream product qualification is claimed.