fix: strip markdown code fences from LLM JSON responses

[Br1an67]

Summary

Some LLM providers (notably Claude Sonnet) wrap JSON responses in markdown code fences (json....) even when JSON mode is requested. This causes json.loads() to fail with JSONDecodeError in generate_schema() and the main extraction methods.

This PR adds a _strip_markdown_json() helper that detects and removes markdown fences before parsing. Applied to:

• generate_schema() (sync)
• agenerate_schema() (async)
• _perform_extraction() sync and async paths

Fixes #1663

List of files changed and why

• crawl4ai/extraction_strategy.py — Added _strip_markdown_json() helper and applied it to all json.loads() calls on LLM responses

How Has This Been Tested?

Verified locally with test cases covering:

• json\n{...}\n (json-tagged fences)
• \n{...}\n (untagged fences)
• Plain JSON (no fences — passed through unchanged)
• Array responses wrapped in fences

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• I have added/updated unit tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes