docs: Revise publication roadmap with rigorous academic standards (#973)

Complete rewrite of publication-roadmap.md from the perspective of a
skeptical top-venue reviewer. Key changes:

- Honest assessment of current evidence (n=45 macOS, n=8 WAA incomplete)
- Acknowledgment that all 45 tasks share same first action (click Apple menu)
- Clear novelty analysis: demo-conditioning is prompting strategy, not architecture
- Anticipated reviewer criticisms with severity ratings
- Required experiments: 240 runs minimum for workshop, ~1500 for full paper
- Statistical rigor: McNemar's test, bootstrap CIs, effect sizes
- 18 essential citations across GUI agents, PbD, VLMs, and RAG
- Realistic venue assessment: workshop 60-70%, NeurIPS main <20%
- 8-week timeline for workshop paper, 6 months for CHI full paper
- Risk mitigation including pivot strategies if results disappoint
- Cost estimates: $200-400 API, $25 compute
- Reviewer response templates for common criticisms

The goal is a paper that could be accepted at a top venue - not just submitted.

Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>

Author: abrichr