Preserve heterogeneous document columns

[codeforester]

Summary

• Derive document columns from the first-seen union of all records.
• Preserve later-only fields in CSV and TSV output.
• Add regression coverage for heterogeneous report records.

Why

When columns was omitted, only the first record determined the output shape. Optional fields appearing in later records were silently dropped.

Validation

• pytest -q tests/test_output.py -k 'render_document'
• full pytest -q
• strict mypy
• Ruff lint
• documentation validation
• git diff --check

Closes #146