DRB-12: Classify missing stage-1 and S1-S11 baseline/scorecard artifacts

[metaphorics]

Problem

Several expected performance baselines and scorecards are referenced by live policy and manifest but do not exist in the working tree. They must be classified as required, intentionally fail-closed, or abandoned, with an explicit owner and lifecycle for each.

Exact missing artifacts

• bench/baselines/stage1.json: expected by bench/compiler-rules.toml:139 and the default in bench/stage1_regression_guard.rs:24.
• bench/scorecards/stage1.json: expected by the default in bench/stage1_regression_guard.rs:28.
• perf/baselines/s1.json through perf/baselines/s11.json: expected by perf/benchmarks.toml:26,36,46,56,66,76,86,96,106,116,127 for slices s1..s11. Only perf/baselines/s0.json is present; s11 is selected = true and is exercised by crates/bamts-verification/src/perf.rs:2620 perf_budget_rejects_slice_without_baseline.

Scope of change

• For each missing path, classify as: (a) required and must be materialized/captured from an approved measurement, (b) intentionally fail-closed placeholder, or (c) abandoned and references removed.
• Update the manifest, gate ledger, and consumers (bench/stage1_regression_guard.rs, perf_budget, bamts-verification perf_guard/tests) to match the classification.
• No fabricated baseline: any newly created baseline must come from an approved measurement run with host conditions and evidence recorded.

Source-of-truth predecessors

• Performance architecture decision: DRB-8 (which decides whether bench/ and perf/ are coexisting or unified into one surface).
• Baseline policy and host conditions: the surviving bench/compiler-rules.toml or perf/hosts/bh1.toml (after DRB-8).
• Scorecard generation: bamts-verification perf_guard::evaluate_stage1_guard_from_paths and perf_budget compare.
• Manifest/benchmark inventory: perf/benchmarks.toml and verification/catalog-inputs.json (benchmarks catalog).

Acceptance

• Every path in bench/compiler-rules.toml, bench/stage1_regression_guard.rs, perf/benchmarks.toml, and perf_guard.rs that names a baseline/scorecard has a documented owner and lifecycle (required/fail-closed/removed).
• No expected = "perf/baselines/s*.json" or baseline = "bench/baselines/*.json" points at a hand-fabricated or unmeasured file.
• perf_budget compare and bamts-verification completion verify --leaf E5.4 produce a deterministic failure mode (missing-baseline error or pass) rather than an unresolved file-not-found panic.
• The manifest or gate ledger explicitly records which S1-S11 slices are selected and which are intentionally unmeasured.

Blockers

• DRB-8 (dual performance authority) must decide whether bench/ and perf/ are the active surface(s).
• DRB-6 (generated cache/workspace residue) may purge ignored boundary fixtures but does not own baseline lifecycle.

[metaphorics]

Dependencies

This issue is blocked by:

• DRB-8: Unify bench/ E5 and perf/ schema-v2 performance authority #196 — DRB-8: Unify bench/ E5 and perf/ schema-v2 performance authority (must decide whether bench/ and perf/ are the active surface(s))
• Purge regenerable ignored cache/worktree residue under recovery rules #193 — Purge regenerable ignored cache/worktree residue under recovery rules (DRB-6, minor — may purge ignored boundary fixtures but does not own baseline lifecycle)

Parent: #156. Feeds: #185.