fix(fetch): pass the correct baseurl to convert_to_md, fix undefined document in use_soup path

[Kumaresan-AI-Engineer]

Summary

• FetchNode's Chromium-loader path called convert_to_md(document[0].page_content, parsed_content) with the same HTML value for both the html and baseurl arguments, so html2text prepended the entire document to every relative link's href, inflating model input by ~74x on pages with many relative links. Now passes the actual source URL as baseurl.
• The use_soup=True path had two related bugs in the same call: the html/url arguments were swapped, and parsed_content was referenced before assignment whenever cut=True (the default in every graph), raising UnboundLocalError.
• The use_soup=True path also never assigned the document variable that handle_web_source unconditionally writes to state, so that path crashed with UnboundLocalError on every request, success or failure. Fixed by assigning document in both the success and failure branches.

Fixes #1142

Test plan

• Added test_fetch_html_convert_to_md_uses_source_as_baseurl — asserts convert_to_md is called with (html, source) on the Chromium path.
• Added test_fetch_html_use_soup_with_default_cut_does_not_raise — exercises use_soup=True with the default cut=True and asserts it no longer raises and calls convert_to_md with correct argument order.
• uv run pytest tests/nodes/fetch_node_test.py passes.
• uv run ruff check, uv run black --check, uv run isort --check-only pass on both changed files.