fix: use crawl4ai result.markdown instead of removed markdown_v2

[obchain]

What

Stop reading the legacy result.markdown_v2 attribute in WebScraper.extract and use result.markdown (the current MarkdownGenerationResult) instead, with a getattr fallback to markdown_v2 for older crawl4ai builds.

Why

crawl4ai 0.5.x removed markdown_v2 and replaced its __getattr__ with a hard AttributeError:

AttributeError: The 'markdown_v2' attribute is deprecated and has been removed.
Please use 'markdown' instead, which now returns a MarkdownGenerationResult

Two call sites in context_scraping/crawl4ai_scraper.py were hitting it:

1. The no_extraction / cosine branch guards with hasattr(result, 'markdown_v2'). On 0.5.x that returns False, so content silently stays None and the scraper returns an empty payload — exactly what About results of the WebScraper #34 reports (Debug: Processed content: None).
2. A few lines later, len(result.markdown_v2.raw_markdown) is unconditional, so on success the scraper re-raises the same AttributeError and the extraction loop falls into the broad except block.

Closes #34

How

src/opendeepsearch/context_scraping/crawl4ai_scraper.py:

• compute markdown_obj = getattr(result, 'markdown', None) or getattr(result, 'markdown_v2', None) once per result
• read raw_markdown from markdown_obj for no_extraction / cosine strategies
• guard the raw_markdown_length / citations_markdown_length bookkeeping on markdown_obj is not None and use getattr(..., '', '') or '' for the individual fields, so a partially-populated MarkdownGenerationResult does not crash the bookkeeping either

The getattr fallback to markdown_v2 keeps the path working for anyone still on a pre-0.5 crawl4ai build (the project's pinned crawl4ai @ git+...salzubi401/crawl4ai.git@main is left untouched on purpose).

Testing

• python3 -m py_compile src/opendeepsearch/context_scraping/crawl4ai_scraper.py — clean
• grep -n markdown_v2 src/opendeepsearch/context_scraping/crawl4ai_scraper.py — only the intentional fallback line remains
• Manual repro path (matches the issue): with crawl4ai 0.5.x installed and WebScraper(debug=True).scrape(<any URL>), the unpatched code logs Debug: Processed content: None and surfaces AttributeError: 'markdown_v2' attribute is deprecated. With this patch, result.markdown.raw_markdown is read instead and content is populated.

No new unit tests added — WebScraper instantiates AsyncWebCrawler and reaches the network, so a meaningful test would need either VCR fixtures or a fake result double. Happy to add a small unit test for _pick_markdown_object if you'd prefer to extract a helper.