fix(openai): correct web citation grounding metadata

[cpsievert]

Why this matters

OpenAI web-search annotations locate inline URL citation markers, not the answer claims that the sources support. This change prevents chatlas from exposing a Markdown source link as ContentCitation.grounded_span.

What changes

• Preserve OpenAI citation URLs, titles, and raw annotation metadata while leaving grounded_span unset in streaming and finalized turns.
• Strengthen VCR coverage around the actual marker offsets.

Verification

• uv run pytest tests/test_provider_openai.py -k web_search -q
• uv run ruff check chatlas/_provider_openai.py tests/test_provider_openai.py

[Copilot]

Pull request overview

This PR corrects how OpenAI web-search URL citation annotations are surfaced by treating OpenAI’s start_index/end_index as offsets for the inline citation marker (not the grounded answer claim), and therefore leaving ContentCitation.grounded_span unset while preserving raw annotation metadata.

Changes:

• Stop deriving grounded_span from OpenAI URL citation annotation offsets on finalized turns; keep the raw annotation payload in ContentCitation.extra.
• Clarify streaming behavior/comments to reflect that OpenAI offsets delimit the inline marker, not grounded answer text.
• Update VCR tests to validate marker offsets/URL presence via extra, and assert grounded_span remains None for OpenAI web-search citations (streaming and non-streaming).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
chatlas/_provider_openai.py	Preserves OpenAI URL citation metadata in extra and stops populating grounded_span from URL citation offsets.
tests/test_provider_openai.py	Adjusts web-search citation assertions to validate marker offsets via extra and ensures grounded_span is unset.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.