Args:
    query: The search query string. Required.
    categories: SearXNG category list, e.g. ["general"], ["images"].
        Defaults to the server's `SEARXNG_DEFAULT_CATEGORY`.
    engines: Restrict to specific engines, e.g. ["google", "bing"].
    language: Language code, e.g. "en", "de". Defaults to server setting.
    pageno: Page number (1-indexed). Default 1.
    time_range: One of "day", "month", "year", or None.
    safesearch: 0 (none), 1 (moderate), 2 (strict). Defaults to server setting.
    max_results: Cap on results returned (1-50). Default 10.

Returns:
    Dict with `results` (list of {title, url, snippet, engine, score}),
    `suggestions`, `number_of_results`, and on error an `error` field.

Args:
    url: An http:// or https:// URL.

Returns:
    Dict with `url`, `final_url`, `content_type`, `text`, `byte_count`,
    and on error an `error` field with a stable string code.

Tries the blog's RSS/Atom feed first (common paths like /rss.xml,
/feed, /atom.xml). If no feed responds, crawls the HTML index
(up to blog_scrape_discovery_pages pages). Each discovered post
URL is fetched in parallel and its cleaned main text is appended
to one Markdown file at:

  {blog_scrape_output_dir}/<safe-host>_<safe-path>_<unix-ts>.md

The file starts with YAML front matter (blog URL, post count,
byte total, duration) followed by one H3 section per post with
its title, original URL, published date (if known from the feed),
content-type, byte count, and the cleaned text. Failures are
noted in-band as `- Fetch error: <code>` bullets, not by aborting
the whole run.

Args:
    blog_url: Root URL of the blog (e.g. "https://blog.example.com/").

Returns:
    Small summary dict — never the file contents:
      {blog_url, output_path, post_count, post_count_ok, byte_count,
       duration_s, discovery}
    On failure: {blog_url, error: <stable_code>, message}.

Same discovery as `scrape_blog` (RSS/Atom first, HTML index
fallback). For every post fetched, the raw HTML is scanned for
links whose URL ends in a recognized document extension. Each
such document is downloaded (capped by
`deep_scrape_max_documents_per_blog`, 100 by default) and its
text is extracted via markitdown. The original blog post text
and every document's extracted text are all inlined into one
Markdown file at:

  {blog_scrape_output_dir}/<safe-host>_<safe-path>_<unix-ts>_deep.md

Downloaded binaries are saved alongside, under:

  {blog_scrape_output_dir}/deep_assets/

Per-post and per-document failures are noted in-band as
`- Fetch error: <code>` / `- Error: <code>` bullets, not by
aborting the whole run. The returned summary dict only carries
an `error` field when discovery itself failed (no posts).

Args:
    blog_url: Root URL of the blog (e.g. "https://blog.example.com/").

Returns:
    Small summary dict — never the file contents:
      {blog_url, output_path, assets_dir, post_count, post_count_ok,
       documents_found, documents_extracted, post_byte_count,
       document_byte_count, duration_s, discovery}
    On failure: {blog_url, error: <stable_code>, message}.