Add ExtractBench Pro leaderboard and result submissions

[xujustinj]

ExtractBench Pro has published scores but no leaderboard CSV or documented way to submit an update in this repository. This adds a Pro table directly alongside the original ExtractBench tables and gives contributors a result-submission workflow.

• Add leaderboard-pro.csv with the twelve published launch systems: overall results and all six challenge slices, using the 30-minute limit. Include perfect-document percentage, Value F1, box/page grounding F1, scoring denominators, and cost per page.
• Document the CSV format, metric definitions, launch snapshot, and evidence needed for new or updated results. Add a benchmark-result PR template.
• Generate both benchmarks' README tables with the existing script and check synchronization in CI. Original ExtractBench scores and generated tables are unchanged.

Validation

• Python 3.12: uv run pytest tests/ -q — 902 passed, 29 skipped (optional runner dependencies).
• uv run ruff format --check and uv run ruff check.
• uv run python scripts/update_readme.py --check and git diff --check.
• Compared every Pro CSV score and cost against the published launch results; verified twelve complete systems, seven slices per system, and the 100-document / 81-grounding-document overall denominators.
• Verified the original CSV and both original generated README blocks are unchanged.

Three new tests catch challenge rows entering the overall ranking or incorrect tie-breaking, unreported grounding/cost being displayed as zero, and stale generated tables passing the check or being overwritten in check mode. No model inference was run.