golden chain: benchmark gc_land against the GitHub loop on the last 80 merged PRs, as a replay spec #9088

The actors rule (see #7851) replaces a loop only after an MVP, its tests and a benchmark against the loop on the same input, with the numbers posted. The golden chain MVP landed as specs/queen/golden_chain.t27 (see #9027, #9029) without that benchmark; its own header says so.

This task replays the last 80 pull requests merged into master through the generated gc_land and compares its answers with what the same pull requests cost in the GitHub loop (update merges, force pushes, conflict resolutions). The input is frozen as a generated replay spec, in the style of specs/queen/replay/actors_decisions.t27, so CI re-checks the benchmark claim on every run.

User Scenarios

• Given the 80 pull requests merged into master up to a fixed master sha, each as one link (merge order, parent = its merge-base mapped to the last window PR merged at or before it, edits = its diff hunks over the base side, derived paths per ci/derived_data.t27 and is_gen, imports = the use targets of the specs it touched), When gc_land from queen/golden_chain.t27 lands them in one pass, Then every outcome count and every per-PR outcome equals what the t27c gen-c output printed for the same arrays.
• Given one real edit region moved so that it meets a region that landed before it, When the same landing runs, Then that link is CONFLICT (negative control).
• Given the GitHub history of the same pull requests, When update merges, force pushes and conflicted merges are counted, Then the numbers are posted on Queen moves to the actor model: the rule in every AGENTS.md, an MVP, a benchmark, then the swap #7851 and Golden chain: the pipeline's own version control -- land on verdicts, not on a queue (MVP) #9027 next to the golden chain's.

Requirements

• FR-001: The replay spec MUST be generated from git and gh data, and its header MUST name the window, the master sha and how to regenerate it.
• FR-002: The landing decisions MUST come from queen/golden_chain.t27 only (use queen::golden_chain); the replay holds data and asserts, no decision logic.
• FR-003: The tests MUST assert the outcome counts and the per-PR outcomes exactly as the generated C printed them, and MUST include at least one negative control.
• FR-004: t27c test-report MUST show 0 FAIL and 0 vacuous for the replay spec, and the spec MUST be sealed with t27c seal --save.
• FR-005: The file MUST be ASCII only (L3).

Success Criteria

• t27c test-report on specs/queen/replay/golden_chain_history.t27 exits 0 with 0 FAIL and 0 vacuous tests.
• .trinity/seals/replay_QueenReplayGoldenChainHistory.json exists and matches the spec.
• The replay holds 80 links and its outcome counts sum to 80.
• A comment with the benchmark table is posted on Queen moves to the actor model: the rule in every AGENTS.md, an MVP, a benchmark, then the swap #7851 and on Golden chain: the pipeline's own version control -- land on verdicts, not on a queue (MVP) #9027.

Boundary

• specs/queen/replay/golden_chain_history.t27
• .trinity/seals/replay_QueenReplayGoldenChainHistory.json