Add CXBench to the Research page

[BoxiYu]

What changed

• Research page: CXBench is now the first project card under Testing AI systems, with approach, findings, and links to the project page, paper PDF, and code. The theme description now also covers process-level checks of agents.
• Publications: new yu2026cxbench entry in _bibliography/papers.bib (marked Preprint, topic Trustworthy AI), so the card's "Full citation" link resolves.
• News: _news/announcement_13.md (2026-10-18) announces CXBench and links to /research/#cxbench.

Why

CXBench (https://cxbench.github.io/) is the benchmark built with Genesys for process-compliant LLM agents in regulated customer-experience workflows. It was not on the site yet.

Notes for review

• Authors, affiliations, and figures (120 tasks, 7 models, 55.5% false-pass, 80.1% of outcome-passing runs, 13.8% both-pass) come from the paper PDF on the project site.
• There's no arXiv ID yet, so the venue reads "Preprint" and the paper link points to the PDF on the project site. Update it once the paper is on arXiv or accepted.
• The homepage Research map (src/data/research.ts) is unchanged.
• npm run build, astro check (0 errors), and npm test (6/6) pass. I checked the generated HTML but not the rendered pages in a browser.

🤖 Generated with Claude Code