Fix mixed sitemap lastmod parsing

[ziodave]

Why

LendingTree's sitemap legally mixes W3C date-only values such as 2026-09-22
with full timestamps such as 2024-10-24T12:20:57+00:00. The SDK parsed the
column without an explicit mixed-format policy and then assumed every value
supported .to_pydatetime(). This caused sitemap ingestion to fail before all
URLs were processed.

Failing production run:
https://github.com/graph-sync/graph-sync-lendingtree-com/actions/runs/36007252436

Sitemap protocol:
https://www.sitemaps.org/protocol.html

What changed

• Parse sitemap lastmod values with format="mixed" so Pandas handles each
  W3C representation independently.
• Normalize parsed values with utc=True, yielding consistent timezone-aware
  timestamps before conversion to Python datetimes.
• Apply the same behavior to both SitemapSourceAdapter and the public legacy
  SitemapUrlSource.
• Add focused regression tests using the exact LendingTree date formats.

Important files

• wordlift_sdk/ingestion/sources.py
• wordlift_sdk/url_source/sitemap_url_source.py
• tests/ingestion/test_sources.py
• tests/url_provider/test_sitemap_url_provider.py

Verification

• poetry run python tests/tools/run_slice_tests.py ingestion -- -q
  • 132 passed, 1 skipped
• Focused source tests
  • 9 passed, 1 skipped
• Affected-module coverage
  • 95% combined
• Ruff check and format check
  • passed
• git diff --check
  • passed

Risks and follow-up

• Date-only values now serialize as midnight UTC with an explicit +00:00
  offset. Offset timestamps are normalized to the equivalent UTC instant.
• Invalid or missing lastmod values remain omitted or None.
• A patch release and a rerun of LendingTree's production_all workflow are
  still required after merge.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.