Resolve relative inline links to fully-qualified URLs during scrape

[arimu1]

Summary

Closes #12

When scraping documentation pages, relative <a href> links in the source HTML (e.g. /self-hosted/latest/migration/entire-database/) were carried verbatim into the generated markdown. Per the issue, this produces broken-looking inline links when an LLM quotes the returned markdown chunks, since the link target isn't resolvable without the source page's base URL.

This resolves relative links against the HTML source (per the issue's suggestion), before the markdown conversion step, across all three ingest pipelines:

• ingest/tiger_docs.py — new resolve_relative_links method on SitemapMarkdownSpider, called in parse() with the current page's response.url as the base, right after the existing strip_data_images step.
• ingest/postgis_docs.py — calls the new shared resolve_relative_links util with each page's resolved full_url as the base.
• ingest/postgres_docs.py — calls the same shared util with a per-page URL built from POSTGRES_BASE_URL, the version, and the page's slug (threaded version into build_markdown, which didn't have it before).
• ingest/utils/beautiful_soup.py — adds the shared resolve_relative_links(soup, base_url) helper (urljoin was already imported here but unused).

Already-absolute links, mailto:/tel: links, etc. are left untouched by urljoin.

Note: this also expands #fragment-only anchors (same-page links) to fully-qualified URLs, since urljoin treats them like any other relative reference. The targets are correct, and it's arguably desirable in a standalone markdown corpus, but PostgreSQL/PostGIS docs are anchor-heavy, so it will visibly increase diff volume in the regenerated output — flagging so it's not a surprise.

Test plan

No Python test harness exists in this repo (ingest/ has no test files and CI (.github/workflows/test.yml) only runs bun test), so I verified behavior manually against the exact example from the issue:

$ cd ingest && PYTHONPATH=.. uv run python - <<'PY'
from bs4 import BeautifulSoup
from markdownify import markdownify
from ingest.utils.beautiful_soup import resolve_relative_links

html = '<main><a href="/self-hosted/latest/migration/entire-database/">migrate your entire database at once</a></main>'
soup = BeautifulSoup(html, "html.parser")
soup = resolve_relative_links(soup, "https://docs.tigerdata.com/self-hosted/latest/migration/some-page/")
print(markdownify(str(soup), heading_style="ATX").strip())
PY
[migrate your entire database at once](https://docs.tigerdata.com/self-hosted/latest/migration/entire-database/)

• Confirmed relative links resolve to fully-qualified URLs, matching the issue's before/after example
• Confirmed already-absolute links and same-page fragment links are handled correctly by urljoin
• uvx ruff check . / uvx ruff format --check . show no new issues introduced by this change (diffed against main, all remaining warnings pre-exist on main)
• python -m ast parses all four changed files without error

🤖 Generated with Claude Code

[arimu1]

Addressed the review feedback:

• Duplicate implementation (blocking): SitemapMarkdownSpider.resolve_relative_links in ingest/tiger_docs.py had its own divergent body (conditional rewrite + debug counter) alongside the shared ingest.utils.beautiful_soup.resolve_relative_links used by the other two pipelines. It's now a thin wrapper that delegates to the shared util (imported from ingest.utils.beautiful_soup) and keeps its debug-level "Resolved N relative links" logging. parse() still calls it with response.url as base, unchanged (ingest/tiger_docs.py:966). Manually verified the wrapper against a mix of relative, fragment-only, and already-absolute hrefs — correct resolution and correct count logged.
• PR body: added a note that #fragment-only anchors also get expanded to fully-qualified URLs (since urljoin treats them like any other relative reference) — correct, but will visibly increase diff volume for the anchor-heavy Postgres/PostGIS docs, so it shouldn't be a surprise.
• No tests added, per the earlier confirmation that this repo has no Python test infra (CI only runs bun test) — verified manually as before.

uvx ruff check . / uvx ruff format --check . show no new issues (same 9 pre-existing warnings as on main, none touching the changed lines).