chore(backlog): T7-32 — 3 micro-investigations bundled (motion_v2 srlv64 + tiny-vmaf-v2 identity + routine.py FIXME)

[lusoris]

Summary

Three S-effort follow-ups identified by the 2026-04-28 BACKLOG audit, bundled in one PR per the audit's hygiene rule.

• (a) motion_v2 AVX2 srlv_epi64 audit. New fork-local libvmaf C unit test libvmaf/test/test_motion_v2_simd.c exercises four adversarial 16-bit fixtures (uniform-negative diffs at bpc 10 and 12; alternating-mixed-sign at bpc 10 and 12) against motion_score_pipeline_16_avx2 in libvmaf/src/feature/x86/motion_v2_avx2.c. The Phase-1 SIMD body uses _mm256_srlv_epi64 (logical) where scalar uses arithmetic >>; the test compares the AVX2 SAD against a line-for-line scalar reference duplicated from integer_motion_v2.c. On the bench host the post-abs() Phase-2 aggregation absorbs the per-lane shift difference and SAD totals match scalar — the test stays as a permanent regression guard. Closes the docs/rebase-notes.md §0038 follow-up placeholder.
• (b) tiny-vmaf-v2 model identity. Research-0006 §4 referenced a non-existent tiny-vmaf-v2 prototype under ai/prototypes/. The actual largest shipped tiny-AI MLP is vmaf_tiny_v1_medium.onnx (mlp_medium, landed by PR feat(ai): tiny-AI training prep (loader + eval + Lightning harness for Netflix corpus) #158). The §4 narrative is updated to reference the real checkpoint name; QAT cost/budget framing unchanged.
• (c) python/vmaf/routine.py:937,1109 FIXME verify. Both cv_on_dataset and explain_model_on_dataset hard-coded feature_option_dict=None with a FIXME comment about inconsistent behaviour with VmafQualityRunner. The FIXME describes a real defect — VmafQualityRunner reads feature_opts_dicts from the model dict at predict time; explain_model_on_dataset did not, so a model carrying per-extractor options would explain itself with mismatched feature configurations. Now: cv_on_dataset reads feature_param.feature_optional_dict when present (mirroring train_test_vmaf_on_dataset at the same file); explain_model_on_dataset reads model.model_dict["feature_opts_dicts"] (mirroring VmafQualityRunner). New regression test python/test/routine_feature_option_dict_test.py covers both None and populated-dict cases for both routines via a FeatureAssembler mock.

Pre-CLAUDE.md §12 r12: no touched-file lint cleanup needed — verify-only sub-tasks.

Deep-dive deliverables (ADR-0108)

• no research digest needed: verify-only sub-tasks, no investigation.
• no alternatives: verify-only fixes (each sub-task has a unique correct answer).
• no rebase-sensitive AGENTS invariants in this PR.
• Reproducer / smoke-test command — see Test plan.
• CHANGELOG.md entry — Unreleased § Changed.
• Rebase note — docs/rebase-notes.md § 0038 closed.

Test plan

• meson test -C build-cpu --no-rebuild — 38/38 OK including new test_motion_v2_simd
• python -m pytest python/test/routine_feature_option_dict_test.py -v --rootdir=python — 4/4 PASS
• pre-commit run --files CHANGELOG.md docs/rebase-notes.md docs/research/0006-tinyai-ptq-accuracy-targets.md libvmaf/test/meson.build python/vmaf/routine.py libvmaf/test/test_motion_v2_simd.c python/test/routine_feature_option_dict_test.py — every hook PASS (clang-format / black / isort / ruff / copyright)
• bash scripts/ci/check-copyright.sh — exit 0
• bash scripts/ci/assertion-density.sh — PASS (every fork-added function ≥20 lines has ≥1 assert)