Fix absolute schema_path handling in Dataset.to_file

[VectorPeak]

This keeps Dataset.to_file() from serializing machine-local absolute schema paths when the schema file is next to the dataset file.

What Problem This Solves

When Dataset.to_file() receives an absolute schema_path in the same directory as the dataset file, it currently compares the schema path against the dataset file path itself:

elif schema_path.is_relative_to(path):
    schema_ref = str(_get_relative_path_reference(schema_path, path))

Sibling files are not relative to each other, so that branch is skipped and the generated dataset stores an absolute $schema reference instead of a portable sibling reference.

Before this change, a local reproduction wrote a machine-local path like:

"$schema": "C:\Users\MiaoXing\AppData\Local\Temp\...\cases_schema.json"

That makes checked-in datasets less portable and can expose local machine paths.

Change

• Compare absolute schema_path against path.parent, the dataset directory.
• Generate the $schema reference relative to that directory when possible.
• Add a regression test for an absolute schema path next to the dataset JSON file.

Evidence

After this change, the same reproduction writes a portable sibling reference:

"$schema": "cases_schema.json"

The schema file is still written to the requested absolute path; only the serialized reference changes.

Validation

• uv run pytest tests/evals/test_dataset.py::test_serialization_to_json tests/evals/test_dataset.py::test_serialization_to_json_with_absolute_schema_path
• uv run ruff check pydantic_evals/pydantic_evals/dataset.py tests/evals/test_dataset.py
• uv run ruff format --check pydantic_evals/pydantic_evals/dataset.py tests/evals/test_dataset.py
• git diff --check

No issue; this is a narrow regression fix for an existing code path.

Checklist

• Any AI generated code has been reviewed line-by-line by the human PR author, who stands by it.
• No breaking changes in accordance with the version policy.
• PR title is fit for the release changelog.