fix: select earliest successful LLM call by lexicographic request_id order

[shawn-mcdonald-dev]

Issue Reference

This PR fixes #2451

Summary

Updated LLM call selection logic to correctly pick the earliest successful call by lexicographic HttpRequestId ordering. This ensures deterministic selection when multiple successful calls exist.

Changes

This PR addresses an issue where the code prioritized completion time instead of lexicographically earliest ulid uuid.

1. Filter to keep only candidates where is_success = true
2. Sort these successful candidates by http request id, ulid uuid, lexicographically)
3. Choose the first candidate from this sorted list

Details

• Filters candidates to only successful calls
• Sorts by request_id.to_string() lexicographically (since HttpRequestId lacks Ord)
• Picks the first entry in the sorted list
• Verified via cargo test -p baml-runtime -q

---

Note

Refactors LLM call selection to pick the earliest successful call by lexicographic HttpRequestId and marks only that call as selected, with new tests covering success-vs-failure and ordering.

• Tracing/FunctionLog (storage.rs)
  • Refactor LLM call assembly: introduce CallCandidate to gather data for each request_id.
  • Selection logic: filter successful candidates, sort by request_id.to_string() lexicographically, select the first; mark only that call as selected across Basic and Stream kinds.
• Tests
  • Add test_selected_call_prefers_success_over_failure and test_selected_call_chooses_earlier_success_if_last_failed to verify selection behavior.

^{Written by Cursor Bugbot for commit 9a3143e. This will update automatically on new commits. Configure here.}

[vercel]

@shawn-mcdonald-dev is attempting to deploy a commit to the Boundary Team on Vercel.

A member of the Team first needs to authorize it.

[hellovai]

@aaronvg can you take a look at this one?