t27c suite Fixed Point: run specs side by side, each spec's two gens still back to back on one worker (#8829 follow-up) #8838

Part of #8737 (hot reload); a follow-up to #8829 / #8837.

Problem

• After t27c suite runs its read-only per-spec subprocess phases in parallel -- those phases 269 s -> 71 s, the suite 498 s -> 276 s (Closes #8829) #8837 the per-spec subprocess phases run in parallel. Phase 6, Fixed Point, still runs t27c gen <spec> twice per spec and compares the outputs, one spec after another: about 3,700 sequential spawns.
• Measured on this Mac: 53-65 s of a 276 s suite. It is now the largest phase after cc_gate, which is the owner's call (not on the exception list).

Change

• A shared helper: the worker pool from t27c suite runs its read-only per-spec subprocess phases in parallel -- those phases 269 s -> 71 s, the suite 498 s -> 276 s (Closes #8829) #8837 moves into par_in_order(items, f). It runs f on every item with suite_jobs() workers and returns the results in input order. run_phase_par uses it.
• Fixed Point uses it too: each spec's two gens stay on one worker, one after the other, so what is compared is still two runs of the same spec. Specs run side by side.
• suite_jobs.t27's comment names Fixed Point among the parallel phases.
• Unchanged: the divergence count, and the first rel_arg error in file order is still the one returned.

Acceptance criteria

• t27c suite --repo-root . --fast --corpus-only --json prints the same Fixed Point: N divergences, the same stderr and the same JSON (apart from timing fields) with T27_SUITE_JOBS=1 and with the default.
• Fixed Point takes at least 2x less wall time with the default.
• Hand-written foreign lines do not grow.

Boundary

specs/verified/suite_jobs.t27

User Scenarios

• Given the test suite runs with default parallelism, When the Fixed Point phase executes, Then multiple specs run their two t27c gen invocations side by side on different workers while each spec's two gens remain consecutive on the same worker.
• Given the test suite runs with T27_SUITE_JOBS=1, When the Fixed Point phase executes, Then specs run sequentially and the divergence count, stderr, and JSON output (excluding timing fields) match the default parallelism run exactly.

Requirements

• FR-001: The worker pool helper par_in_order(items, f) MUST be introduced in suite_jobs.t27 and run f on every item with suite_jobs() workers, returning results in input order.
• FR-002: run_phase_par MUST use par_in_order for parallel phase execution.
• FR-003: The Fixed Point phase MUST use par_in_order so that each spec's two gen runs execute back-to-back on one worker while different specs run concurrently.
• FR-004: specs/verified/suite_jobs.t27's comment MUST name Fixed Point among the parallel phases.
• FR-005: The divergence count and the first rel_arg error in file order MUST remain unchanged between parallel and sequential runs.
• FR-006: Hand-written foreign lines MUST NOT increase as a result of this change.