docs: lead ReflexBench v1 with raw engine lane

[0xnavarro]

Summary

• makes the frozen 111-case same-corpus raw-engine lane the v1 launch headline
• keeps Reflex Core same-response/workflow evidence public as a clearly separate supplemental research lane
• preserves all underlying receipts and methodology; no benchmark data or scoring changes
• aligns README and RESULTS with the public launch narrative: model quality first, policy/harness ablations second

Canonical raw lane

Engine	Raw semantic accuracy
TypeSafe Jev	73.0%
upstream Reflex / Qwen3.5-2B	41.4%
frozen Qwen3.5-0.8B readout control	39.6%
jeff / GLiFormer ~400M	37.8%
openJev Verdict 1.4 / 151M	36.9%
Laya base / 421M	35.1%
Kev-0.8B	32.4%

QA

• PYTHONPATH=reflexbench python3 -m unittest discover -s tests -p 'test_*.py' -v — 50/50 pass
• git diff --check — pass

Docs-only launch framing. No receipt, corpus, runner, scoring or release-tag mutation.