Fix CI: pin ruff, make type inference tests pandas 3 compatible

[chrisclark]

Summary

master CI has been red because of dependency drift, not code changes. This blocks #710 and #687.

Lint

• lint.yml ran pipx run ruff without a pinned version, so new ruff rules started failing on unchanged code. It's now pinned to ruff==0.16.8.
• Added target-version = "py38" to ruff.toml to match python_requires (ruff can't read setup.py). Newer ruff defaults to py310 and suggests X | None, zip(strict=) and similar, which would break 3.8/3.9.
• Fixed the remaining findings:
  • Replaced % formatting with f-strings in get_version() and the user factory. Output is unchanged, checked for final, patch and beta versions on Python 3.8 and 3.10.
  • Added a noqa: PLR0917 on query_viewmodel instead of changing a signature that 5 call sites use.

Tests (pandas 3)

• The dev requirements now resolve to pandas 3.0, which gives datetimes a datetime64[us] dtype instead of [ns], and text a str dtype instead of object. Six test_type_infer tests asserted those exact dtypes.
• I checked the actual upload output. Every CSV/JSON test fixture was run through pandas_to_sqlite on pandas 2.2.2 and 3.0.6, and the resulting SQLite column types and values are identical. Upload behavior hasn't changed.
• The tests now check for any datetime dtype, and for object or string dtype for text columns.

Testing

• Full suite passes locally on pandas 2.2.2 and 3.0.6, under both explorer.tests.settings and settings_base.
• ruff@0.16.8 check explorer is clean.

🤖 Generated with Claude Code