[M1-BENCH-01] laige-bench sim-tick suite: 10k-entity tick budget on both SimMath backends

[offdev]

Implements M1-BENCH-01 (roadmap/M1-heartbeat.md).

What

The PRD §8.1 "Simulation tick" workload as a laige-bench suite (--suite=sim-tick):

• Workload — 10 000 entities at the 100% scene budget, 2 000 dynamic bodies (Position2D<B> + a BenchVel<B> component, both marked deterministic-safe), 8 000 bare entities; deterministic seed 0x1F055EED, index-derived initial state (no RNG in the measured path).
• Systems — BenchMove (1 ms budget, pos += vel via the active backend's SimMath::add) and BenchHash (2 ms budget, the per-tick World::stateHash).
• Driver — the engine's GameLoop at 60 Hz on an exact synthetic clock (one due tick/frame); one measured sample = one completed tick; 1 000 warm-up + 3 000 measured ticks.
• Backends — both SimMath backends (--math=fixed_point_16_16 / float_pinned_32), both must pass (ADR 0002).
• Zero-alloc — in debug non-sanitizer builds, every measured tick additionally passes the engine's own G-R1 per-tick zero-allocation assertion (M1-ALLOC-01); an allocating tick aborts the run.

Tool changes

• --math=<fixed_point_16_16|float_pinned_32> — SimMath backend selection for the sim-tick suite (default fpx16).
• Repeatable --budget=<name> — one run checks every named budget against the same histogram (the M1 gate: avg AND p99 in one run); duplicate name = usage error; any FAIL exits 2.
• Compiler-id fix: Clang is now identified before GCC (Clang 22 defines __GNUC__ and its __VERSION__ is "Clang 22.1.8" — the old order mislabeled clang builds as "GCC Clang 22.1.8").

CI gate

New ctest entries laige_bench_sim_tick_fpx16 / laige_bench_sim_tick_fp32 (canonical 3000/1000 shape, both budgets, repo-root budgets.json via LAIGE_BUDGETS_PATH) run in every non-instrumented P0 job's full ctest — this is the PRD §14 CI perf lane for the sim tick. Excluded from the ASan/TSan trees: instrumentation inflates the tick's absolute cost (~2.7× under ASan+UBSan), so a sanitizer measurement would measure the instrumentation, not the sim (the m1-profiler-cost precedent). A budget FAIL exits 2 and fails the CI job.

Results (canonical Debug, g++ 16.2.1, AMD Ryzen 9 7950X3D / CachyOS)

Backend	mean (ms)	p99 (ms)	Target
fixed_point_16_16	0.59283	0.618314	≤ 3.0 / ≤ 5.0
float_pinned_32	0.589047	0.605951	≤ 3.0 / ≤ 5.0

Both backends PASS (5.1× / 8.1× inside the budgets). The run's final_hash fingerprint is bit-identical across all four non-instrumented local trees and both compilers (g++ 16.2.1 / Clang 22.1.8), at both backends (fpx16 0x8aa6b855aa807b31, fp32 0x34ff9f20dac09426).

budgets.json gains its first recorded measured values (worse of the two backends): sim_tick_avg 0.59283, sim_tick_p99 0.618314.

Baseline + docs

• Fourth baseline: docs/benchmarks/baselines/m1-sim-tick.md — full AGENTS §12 metadata, verbatim budget-checked runs, the ctest gate, cross-tree/cross-compiler table, interpretation. Records that M1 measures the ECS-only slice of the §8.1 workload (kinematic movement + state hash; physics lands in M3, presentation in M4).
• docs/benchmarks/README.md + docs/benchmarks/baselines/README.md — stale "every entry still measured: 0" text replaced; the missing m1-profiler-cost baseline entry added.
• docs/api/budget_harness.md — the new CLI form, the sim-tick suite, the measured-field note, both suites' determinism.

Local verification

Full ctest green on all six local trees: 94/94 build/build-clang/build-shared, 93/93 build-release, 92/92 build-asan/build-tsan (gate entries correctly absent there); ASan run leak-free, TSan halt_on_error=1 race-free (one legitimate G-R5 budget_overrun warn on BenchHash under the 4× TSan slowdown — correct engine observation); zero new warnings under NFR-8.10; include-lint + determinism-lint OK; no new public API (laige-api.json unchanged).