fix(web): add Cloudflare block detection and direct HTTP fallback

[nankingjing]

Summary

When Jina Reader is blocked by Cloudflare or similar anti-bot protection (e.g., fetching imginn.com/instagram/), the response comes back as a security verification page instead of actual content.

This PR adds automatic Cloudflare block detection and a direct HTTP fallback using requests (already a project dependency) with realistic browser headers.

Changes

• agent_reach/channels/web.py: Added _is_blocked() to detect Cloudflare/anti-bot block pages and _direct_fetch() as a fallback using browser-like headers
• When Jina Reader returns a block page, auto-fallback to direct HTTP fetch
• Only changes the web channel; all other channels unaffected

Test Results

184 passed, 11 skipped, 1 pre-existing failure (unrelated YouTube test)

Closes #428

[moxxxskaifa]

Nice fix! 解决了一个长期痛点：

• _CF_BLOCK_PATTERNS 覆盖了常见 CF 拦截页面特征 ✅
• _is_blocked() 只检查前2000字，性能开销小 ✅
• Jina Reader 被拦后自动降级直连，优雅降级 ✅

建议一个小优化：direct_fetch 失败时最好 fallback 回原始 text（而不是丢数据）。

整体 LGTM!

[nankingjing]

您好 @moxxxskaifa，这个 PR 已经 approved，是否可以合并了？谢谢！

[nankingjing]

Hi @moxxxskaifa, this PR has been approved — could you merge it when you have a moment? Thanks!

[nankingjing]

@moxxxskaifa 感谢 review！已按您建议的小优化实现：

当 Jina Reader 被判定为被拦截、走 _direct_fetch 直连回退，而直连本身也失败（返回 None 或空响应）时，read() 现在会显式回退到原始 Jina 文本（direct or text），不再有丢数据的风险；空字符串也会走回退。逻辑与原有降级链保持一致，未改变其它路径。

commit: f0fb2fd

[nankingjing]

@moxxxskaifa the post-approval commit f0fb2fd implements your suggestion exactly: when _direct_fetch fails, read() now explicitly falls back to the original Jina text (direct or text) instead of losing data. No other paths changed. The earlier approval was on 39eecdc — would you mind confirming this still LGTM? Thanks!

[Panniantong]

感谢定位 Cloudflare challenge 页面。merged #576 已安全检测并明确失败；本 PR 的直接 HTTP fallback 可能绕过预期网络和安全边界，因此不采用并关闭。