ci(build): retire the completed comparison harness and preserve its method

[DecisionNerd]

Native completion blocker of #1618. The existing agent retains #1617/#1618 implementation ownership; this session coordinates integration only. Start from current main after #1646 merges. This issue closes the historical measurement-method and temporary-harness retirement gap; it does not reopen the Cargo decision or expand product scope. The separate differential-oracle selection correction is owned by #1662.

Verified remaining outcomes

The build-tree retirement PR #1661 satisfies its #1646 concern, but two canonical outcomes remain:

The now-completed dispatch-only comparison harness still exists with gate-registry entries and artifact/cache/sticky policy contributions. ADR0048 points to that live temporary harness and does not record the producing source/input digests. The completed twenty-record cohort and maintenance results already live on #1618 comment5902091518. Method and input hashes belong in the linked repository topic; no measurement rerun is needed.

Scope

• Fold only the historical measurement method and immutable producing source/input digests into linked testing/agent-environment documentation; update ADR0048's evidence/revisit pointers. Keep raw results on decision(build): make one build system the CI authority; measure Bazel remote cache against Cargo with sccache #1618. Read-only input materials: /tmp/graphforge-1646-retirement-plan.i8w37n_7/measurement-method-draft.md, measurement-input-digests.json, and /tmp/graphforge-1618-evidence-audit.fwK6Nc.
• Retire .github/workflows/build-lane-measurement.yml, its registry command/supporting-workflow records, related storage-policy accounting and inactive documentation references together. No replacement permanent gate/harness.

Acceptance criteria

• The completed temporary workflow and registry/policy contributions are removed consistently; workflow/storage policy and make gate-registry-check pass, and CI Gate remains the sole required status.
• A linked topic records reproducible historical commands, runner/cache semantics and content digests identifying inputs. ADR evidence pointers work after retirement. Raw tables, JSON and logs remain on decision(build): make one build system the CI authority; measure Bazel remote cache against Cargo with sccache #1618.
• Exact-head CI Gate and the merge queue pass before merge. Post resulting commands/results on this issue and retain decision(build): make one build system the CI authority; measure Bazel remote cache against Cargo with sccache #1618 as the canonical close gate.

Boundaries

No Rust product behavior or format changes, no new release certification cascade, no new benchmark cohort, no differential-oracle selection changes (owned by #1662), and no storage-policy implementation from #1617. Retired assembler map parity is outside scope; the prior P1 request to retain its fixed nineteen-entry map was independently challenged and withdrawn in #1661.

[DecisionNerd]

Integration-only independent review of your clean 257b0ba9 head found no blocker. Registry/storage/workflow/docs checks pass; all eight historical workflow source hashes, twenty measured Cargo.lock hashes, the cohort digest, and ten main-history ancestry claims reproduce. Your branch ownership remains unchanged.

Two optional method clarifications before publishing: identify the pinned historical workflow timed block as authoritative for historical commands (the linked current commands now use test_environment.py); briefly distinguish cold target, sticky target, and sccache disks from the separate Cargo dependency cache. The pinned source already preserves these inputs, so neither calls for another measurement cohort.

Please keep the PR limited to #1663. I will coordinate frozen-head review and the merge queue once it is open.