[M1-ECS-07] ECS stress + memory accounting test

[offdev]

Implements roadmap step M1-ECS-07 · ECS stress + memory accounting test (scope only; no other changes).

What

New EcsStress suite in laige-sim_tests (CTest entry ecs_stress, added to the TSan property list):

• 10k entities at the 100% scene budget (capacity 10000), 6 component types (4/8/8/16/8/4 B — distinct column strides), 10k frames of add/remove churn.
• Each frame: beginFrame() drives G-R3/G-R4, then 64 seeded Tag adds + 64 Tag removes (cyclic Fisher-Yates permutation, TestPrng substream 1008; 128 ops/frame — the default 256 G-R4 budget is never exceeded), then one each<Pos, Tag> iteration (Read, Read) with visit count + 64-bit FNV-1a checksum.
• A 700-frame warm-up (one full 625-frame cohort period + margin) brings every archetype's columns to their high water before the window, so the window's zero reservation/growth delta is the "pool high-water stable" claim (measured high water: 9 archetypes, 24,592 reserved rows, 549,024 bytes).

Verified properties (the step's Scope)

Property	Check
No leaks	ctest -R ecs_stress green on build-asan (37.7 s, no ASan/UBSan report)
Pool high-water stable	zero totalReservations/totalArchetypeGrowth delta across the 10k-frame window
Iteration within documented cost (query.h)	window-wide ns/visit throughput floor: 600 ns; measured 58.7 ns (g++ 16.2.1, -O0 Debug) / 72.3 ns (clang++ 22.1.8)
Zero-allocation window	test-only operator-new counter (== 0); sanitizer trees: leak-free run + delta
Memory accounting (PRD §8.1 base memory)	110,000 entity bytes + 549,024 reserved row bytes; machine-greppable ecs-stress window/iteration/memory lines on every ctest run

The window/memory lines are byte-identical across g++ and clang++ (pure integer workload, ARCH-010) and across repeated runs. No new guardrail warns in the window (churnWarns delta 0; the entity-budget warns fire once at setup and never re-cross).

Baseline

Second baseline file: docs/benchmarks/baselines/m1-ecs-stress.md — full AGENTS §12 metadata, verbatim runs (canonical, ASan, cross-compiler), cross-tree results. Not a budgets.json workload, so no measured field is updated. Baselines index + benchmarks README updated to match.

Local verification

• ctest -R ecs_stress green on build (Debug g++, 14.7 s), build-asan (required Verify), build-release, build-clang, build-tsan, build-shared
• Full suite 40/40 on build, build-asan, build-clang; laige-sim_tests + ecs_stress green on the other trees
• Zero new warnings under NFR-8.10 on all trees
• No public API added (laige-api.json unchanged, 452 symbols — api-real-tree green); no include-graph change

Note

This PR also adds the retroactive M1-ECS-06 row to the roadmap change log (and the matching board counts, M1 5→7 / total 27→29): M1-ECS-06 merged (675fb94, #22) without updating roadmap/README.md. Same pattern as the M0 retro rows.