ci: try to make tests more stable

[Xuanwo]

This PR will try to make tests more stable

---

Parts of this PR were drafted with assistance from Codex (with gpt-5.2) and fully reviewed and edited by me. I take full responsibility for all changes.

[github-actions]

PR Review

Summary: Test stability improvements for FTS result comparison + uv lockfile refresh.

P1 Issues

1. Potential bug in sort fallback logic (test_scalar_index.py:2409-2411)

When _rowid is not present, the fallback sorts by all columns except _score. However, if the DataFrame has only a _score column (unlikely but possible), sort_cols becomes an empty list. While the if sort_cols: guard prevents an error, it means no sorting happens, which could cause test flakiness in edge cases.

2. dtype=object conversion may mask type mismatches (test_scalar_index.py:2446-2449)

Converting all non-score columns to dtype=object before comparison is convenient but loses type information. The original code handled object vs non-object columns differently. If there's a legitimate type mismatch between single-machine and distributed results, this change would silently ignore it.

Consider keeping type-aware comparison or at least logging/asserting dtype equality:

assert single_df[col].dtype == distributed_df[col].dtype, f"Column {col} dtype mismatch"

Minor Observations

• The lockfile changes are routine (uv adding upload-time metadata). The large diff is just metadata, no package version changes.
• The kind="mergesort" choice is correct for stable sorting.

Overall the test stability improvements are reasonable, just the type-handling concern above warrants consideration.

[jackye1995]

looks good to me!