docs: add Quick Start & CHANGELOG, embed prepared benchmark charts, asset cleanup

[SuyangLiuPaul]

Summary | 改动摘要

Documentation improvements, applied to both README.md and README_CN.md in sync. 文档改进，中英文 README 已同步修改。

• 🚀 Quick Start section near the top (+ nav bar entry): Homebrew install, cloud/local run examples, and the macOS permission note — visitors no longer need to scroll past the benchmark section to find the install commands.
  新增「快速开始」小节及顶部导航入口：brew 安装、云端/本地运行示例、macOS 权限提示。
• Embedded the four prepared-but-unreferenced benchmark charts (Video-SME-2, MIT1003-SalECI, ETMD, Online-Mind2Web) above their data tables inside the existing <details> blocks, matching the pattern of section 1.
  将 pics/ 中已制作但从未引用的 4 张基准图表嵌入对应折叠块（图在前、数据表在后，与第 1 节模式一致）。
• CHANGELOG.md reconstructed from git history, covering 1.0 (2026-03-25) and 1.1 (2026-06-01).
  新增 CHANGELOG，覆盖 1.0 与 1.1 两个版本。
• Asset cleanup: renamed pics/MIT1003, SalEC.png → pics/MIT1003-SalECI.png (URL-unsafe comma/space; spelling now matches the section title); removed unused mano-action.jpeg (256 KB). Verified the HF model cards hotlink no repo images, so the rename/removal breaks nothing.
  资产清理：重命名含逗号空格的图片文件、删除未使用的 jpeg；已确认 HF 模型卡未热链仓库图片。
• EN hero banner now links to the in-page benchmark section instead of the Feishu wiki (login-gated for most international visitors). The CN version keeps the Feishu link.
  英文版首屏大图改为页内跳转（飞书 wiki 对海外访客需登录）；中文版保留飞书链接。
• Contribution guidelines now state the EN/CN READMEs must be kept in sync.
  贡献指南补充中英文 README 同步维护的约定。

Verification | 验证

• Every image referenced by either README exists; all 13 files in pics/ are now referenced.
• EN/CN structural element counts (headings, images, <details>) match: 81 / 81.
• All external links in both READMEs return HTTP 200.

Suggested maintainer follow-ups | 建议后续（需仓库权限）

1. Tag v1.1 (e.g. at the merge commit) and publish a GitHub Release — draft notes below. 打 v1.1 tag 并发布 Release，说明稿见下。
2. Set the repo homepage (About) to the Hugging Face model page. 在 About 区设置 homepage 链接。

📋 Draft release notes for v1.1 / v1.1 Release 说明稿（可直接使用）

Mano-P 1.1

Model iteration 1.1 — released weights are now published under the standardized name Mano-CUA-4B-Thinking-1.1, including an MLX-8bit variant for fully on-device inference on Apple Silicon.

🤖 Models

Platform	Standard	On-device (MLX 8-bit)
🤗 Hugging Face	Mano-CUA-4B-Thinking-1.1	Mano-CUA-4B-Thinking-1.1-MLX-8bit
🪄 ModelScope CN	Mano-CUA-4B-Thinking-1.1	Mano-CUA-4B-Thinking-1.1-MLX-8bit
🪄 ModelScope AI	Mano-CUA-4B-Thinking-1.1	Mano-CUA-4B-Thinking-1.1-MLX-8bit

📊 Highlights

• #1 on OSWorld among specialized GUI agent models: 58.2% success rate (second place: opencua-72b at 45.0%)
• 41.7 NavEval on WebRetriever Protocol I, ahead of Gemini 2.5 Pro Computer Use (40.9) and Claude 4.5 Computer Use (31.3)
• New Instruct vs Thinking comparison on the 100-task execution suite: Thinking-4B 56.0% vs Instruct-4B 47.0% pass rate (cloud 83.0%, Qwen3-VL-Plus 39.0%)
• New on-device vs cloud vs Qwen3-VL task-execution benchmark with per-tier and per-category breakdowns
• ~80 tokens/s decode on Apple M5 Pro; with Cider W8A8 activation quantization, prefill speeds up ~12.7% over the W8A16 baseline

🚀 Try it

brew tap Mininglamp-AI/tap
brew install mano-cua

# cloud mode (default)
mano-cua run "Open Safari and search for Python"

# fully on-device (macOS Apple Silicon)
mano-cua check && mano-cua install-sdk && mano-cua install-model
mano-cua run "Open Safari and search for Python" --local

🤖 Generated with Claude Code