docs: lead with independent trust harness

[nugehs]

Summary

• Lead Otito with independent merge evidence: models generate the change, Otito proves whether it is safe to merge.
• Add the trust-harness thesis and wire it through the README, docs site, exec summary, glossary, and companion thesis pages.
• Treat native agent loops (Codex, Claude Code, Gemini, Cursor) as complementary infrastructure, not competitors.

Test plan

• Read the README lead and confirm it no longer argues that a tighter harness mainly makes cheaper models smarter.
• Open docs/14-trust-harness-thesis/README.md and the docs index/nav entry.
• Spot-check companion links from dual-mode, determinism, harness, and prompt-determinism pages.
• Confirm AX remains described as a cost/guardrail diagnostic, not the product lead.