feat(v2.6.0): signature-based anti-bot compatibility + MCP v0.5.0 alignment

- Add top-level chapter: anti-bot type classification (signature/behavioral/obfuscation)
- Add observer effect warning for pre_inject_hooks on signature-based anti-bot
- Add hook_jsvmp_interpreter mode parameter (proxy/transparent)
- Update instrument_jsvmp_source AST mode: MCP-side esprima (zero CDN dependency)
- Add fallback_on_error parameter and last_mode_used status field
- Add four-axe applicability labels per anti-bot type
- Rewrite scenario 9 (anti-detection sites) by anti-bot type
- Add scenario 11 signature-type pre-check
- Add observer effect troubleshooting section
- Add experience rules 28-32
- Add AST health diagnostics to jsvmp-source-instrumentation.md
- Add transparent mode examples to mcp-cookbook.md
- Add anti-bot type column to cases/README.md
- Create CHANGELOG_v2.6.0.md

cases/universal-vmp-source-instrumentation.md

@@ -82,6 +82,11 @@ Step 5 — 装兜底 hook（若 Step 2 没走 pre_inject_hooks）
   inject_hook_preset("fetch", persistent=True)
   inject_hook_preset("crypto", persistent=True)
   hook_jsvmp_interpreter(script_url="[VMP basename]")
+  # 注意：本案例针对的是"通用"场景（含签名型和行为型）。如果目标是签名型反爬
+  # （瑞数/Akamai），不要这样用 hook_jsvmp_interpreter，改为：
+  #   - instrument_jsvmp_source(mode="ast")  （首选）
+  #   - hook_jsvmp_interpreter(mode="transparent")  （备选）
+  # 参考 SKILL.md "反爬类型识别与工具选择" 章节
   bypass_debugger_trap()
 
 Step 6 — reload_with_hooks 让探针先于 VMP 生效