fix(testing)!: grade every verifier properly

[sephynox]

Summary

Two harness verifiers were reporting results nobody computed: Layer 3's verdict had no reader since 596d3e1, and the gate check compared each spec against a decision the harness itself had written moments earlier. Both now answer from the evidence they hold, which turned 282 silently-wrong results into real verdicts.

Related Issues

Refs .audit/60-synthesis/REMEDIATION.md phase 6

Changes Made

• Layer 3 reports one decision per assertion, not five booleans
• A recorded trace is checked in the traces model only
• An undecided check is reported as such, not as a violation
• gate: is optional, and the fabricated gate decision is gone
• A run with no gate decision is accepted, not read as an error

Testing

• make test-all: passes

Breaking Changes

• FdrVerdict fields are Decision. passed and complete are gone, use outcome()
• FdrConfig.expect_failure is gone. Use .with_expect(Expect::Violation(layer))
• gate: at a site that takes no gate now fails. Remove the declaration
• Process is #[non_exhaustive]. Construct through Process::builder
• ScenarioResult fields are private. Use trace(), verdict(), passed()

---

Note

Medium Risk
Changes core test verdict semantics and breaks callers that relied on implicit gate: Ok or old FDR/ScenarioResult APIs; incorrect specs may now fail where they previously passed silently.

Overview
Gate expectations are no longer fabricated or mandatory. The scenario harness stops writing TransitStatus::Ok into every consumed trace, and tb_assert_spec! treats gate: as optional—omitting it means the spec requires no gate decision. Specs that still declare gate: Ok are checked against the trace’s real decision; a gateless run now fails Layer 1 assertion verification instead of passing because the harness lied.

Trace semantics align with “no gate taken.” ConsumedTrace::execution_mode() treats missing gate decisions as Accept, not Error, and FDR trace helpers treat no decision as valid/terminal where appropriate instead of requiring a gate for every run.

Mechanical fallout: AssertSpecBuilder::gate_decision becomes expected_gate(Option<TransitStatus>), macro plumbing passes Option::None or Some(status) from optional gate: blocks, and hundreds of tests/fuzz specs drop redundant gate: Ok. New tests/gate scenarios lock in gateless acceptance vs. gate: Ok rejection.

^{Reviewed by Cursor Bugbot for commit 4382258. Bugbot is set up for automated code reviews on this repo. Configure here.}