Publish v0.8 execution evaluation results

[AbdelStark]

Summary

• publish the v0.8 correctness-aware execution results report and public artifact index
• add tracked docs/benchmark/v0_8/ eval manifests/reports/score rows for both seeds
• add v0.8 dataset/model cards with the HumanEval-positive but overall-claim-closed boundary
• add docs regression coverage so future edits keep the v0.8 claim boundary explicit

Job status

• seed 42 HF job 6a2278d2e6aa50b87b9eba56: COMPLETED
• seed 1729 HF job 6a227a6ce52fdd2a02ed9005: COMPLETED
• both uploaded verified run artifacts; checkpoint mirrors were downloaded back and verified

Result boundary

• HumanEval WS-D rerank passes on both seeds
• MBPP-Plus WS-D rerank does not clear the claim gate
• pass/fail latent probe does not beat controls consistently
• output-magnitude probe is not evaluable because the v0.8 pack has zero val labels for that target
• overall v0.8 downstream claim remains closed

Validation

• uv run pytest tests/ -> 939 passed, 8 skipped, 1 warning
• uv run pytest tests/docs -> 122 passed
• uv run python -m compileall -q -x 'tests/fixtures/codestate/invalid_(before|after)\.py$' codelewm tests
• git diff --cached --check
• parent-aware manifest verification for all 10 copied docs/benchmark/v0_8 eval manifests
• uv run codelewm secret-scan docs/benchmark/v0_8 docs/benchmark/EXECUTION_V0_8_RESULTS_2026-06-05.md docs/benchmark/PUBLIC_ARTIFACT_INDEX_2026-06-05.md docs/cards/codelewm-v0-8-execution-dataset-2026-06-05.md docs/cards/codelewm-v0-8-execution-model-seed-42-2026-06-05.md docs/cards/codelewm-v0-8-execution-model-seed-1729-2026-06-05.md --json

Closes #371.