Fix CSV export for empty result sets

[GHX5T-SOL]

Summary

• Fix CSV export when every model call fails and export() receives no records.
• Use the exporter schema as the CSV field list instead of deriving headers from records[0].
• Add regression coverage for empty CSV exports writing the header row and no data rows.

Fixes #8.

Validation

• python3 -m pytest tests/test_exporter.py -q -> 8 passed, 1 warning
• uv run --with pytest --with pytest-asyncio --with ruff --with numpy pytest -q -> 40 passed, 4 skipped
• uv run --with pytest --with pytest-asyncio --with ruff --with numpy ruff check . -> All checks passed
• git diff --check
• git diff | gitleaks stdin --no-banner --redact --timeout 30 -> no leaks found
• Direct smoke: export([], temp_csv) writes only the CSV header row

Notes

I could not use the declared full dev extra on this host because it resolves to torch==2.12.0, which has no compatible macOS x86_64 wheel here. I used the minimal test/lint dependencies needed for the local test suite instead.

[GHX5T-SOL]

Closing this as duplicate/stale. PR #10 was merged earlier and resolves #8, so this branch is no longer needed.