[M1-ECS-05] Deterministic iteration order

[offdev]

M1-ECS-05 from roadmap/M1-heartbeat.md: the documented deterministic iteration contract over the World::each visit order, plus the convergence property test.

What changed

• Contract documented (docs/api/iteration_order.md): archetypes in ascending archetype id (= the order a component set is first seen by a mutation — creation order), entities within an archetype in ascending slot id, empty query ascending slot id. The order is a pure function of the world state, never of the operation history.
• Dense-id-order scheme documented: rows stay in ascending slot-id order through insert/remove (archetype.h invariants I1–I4), so two histories converging on the same state visit identically.
• No unordered containers in the iteration path: only the fixed 256-record archetype table (id-ordered scan), packed slot columns, per-slot direct-index records, and membership-only 256-bit guard sets. The anticipated "one internal hash structure" (entity→archetype) is direct indexing — not even a hash, a stricter reading of the step allowance; the one hash structure in laige-sim (the component type-key index) is lookup-only, never iterated, and setup-path only.
• Convergence property test (tests/laige-sim/iter_order_tests.cpp, CTest iter_order): two worlds whose operation sequences interleave create/destroy differently — A: serial per-entity scripts + end-phase dead destroys; B: the same create phase (LIFO requires it for the identical entity→id assignment) with PRNG-scattered dead-destroys, round-robin component steps in per-round PRNG permutations, and PRNG-placed scratch pairs — converge on the identical final state including the entity→id assignment and iterate identically for five queries, verified per-visit against an independent oracle. Component moves are structural throughout, pinning the dense-id scheme. A KAT pins the exact degenerate visit sequences; the greppable iter-order … fnv1a=0x… line is byte-identical across all instantiations, trees, and seeds.
• Docs cross-refs (query.md, entity.md, archetype.md, component_registry.md, docs/README, sim README); roadmap checkbox + progress board (M1 5/25, total 27/193); change-log row in commit 2.
• No public API added — laige-api.json unchanged (452 symbols, scanner rerun clean); no include-graph change; include lint OK.

Verification

• ctest -R iter_order green (the step Verify command): build (g++), build-asan, build-tsan, build-clang, build-release
• Full laige-sim_tests suite green on build (no regressions)
• Visit hash 0x12fc225187f3b359 byte-identical across g++/Clang/ASan/release trees and across LAIGE_TEST_SEED overrides
• Size: the step estimated ~100 lines + tests; the test suite is ~960 lines (the convergence construction + oracle dominate — same order as M1-ECS-04, whose ~250-line estimate landed at ~1450)

Untested: no new runtime system, so no benchmark/diagnostics scope; the no-unordered-containers half is verified by inspection (CI source scan lands in M1-DET-01).