Add ExtractBench Pro download and evaluation support

[xujustinj]

The public harness only downloads the original short/medium/long dataset. Add --dataset pro to download, status, and run so users can download all six Pro challenge splits and evaluate them with the existing scorer. The README includes copyable download, evaluation, and report commands.

Pro defaults to data-pro/ and output-pro/, keeping it separate from original ExtractBench. Downloads preserve the resolved Hugging Face revision, reject a directory owned by the other dataset, and repair interrupted refreshes without accepting stale PDFs as ready. Pro has no --test subset; --group selects one challenge.

Validation:

• Ruff format and lint.
• Full runner-free suite: 899 passed, 29 skipped for optional dependencies.
• Built wheel and sdist with uv build.
• Downloaded and loaded the actual public Pro dataset: 100 documents across all six challenges. Dataset lint reported zero findings.
• Local scoring smoke test using ground truth as predictions for one real document per challenge: all six scored Value F1 = 1. No paid inference was run.
• Verified an interrupted-refresh regression fails before the fix and passes afterward: retry must replace an existing stale PDF.
• git diff --check.

New regression/contract tests cover: public CLI reconstruction into loader-readable cases and missing PDFs; cross-dataset overwrite rejection in both directions; invalid selection before download/inference; routing both inference and evaluation into Pro directories; refreshed PDFs and removed documents; and interrupted refresh readiness/retry behavior.

[xujustinj]

Looks like it works!

xujustinj@Justins-MacBook-Pro ExtractBench % uv run extract-bench download --dataset pro
Downloading dataset from HuggingFace: llamaindex/ExtractBench
Destination: data
Downloading (incomplete total...): 0.00B [00:00, ?B/s]                                                                 Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
Fetching 402 files: 100%|█████████████████████████████████████████████████████████████| 402/402 [00:08<00:00, 46.10it/s]
Download complete: : 258MB [00:10, 24.3MB/s]                                                                            
Dataset ready at: data (370 test cases, revision 51bf3a03eb7f)
ERROR: Could not consume arg: --dataset
Usage: extract-bench download - <command|value>
  available commands:    as_integer_ratio | bit_count | bit_length |
                         conjugate | from_bytes | is_integer | to_bytes
  available values:      denominator | imag | numerator | real

For detailed information on this command, run:
  extract-bench download - --help
xujustinj@Justins-MacBook-Pro ExtractBench % git checkout codex/pro-download
branch 'codex/pro-download' set up to track 'origin/codex/pro-download'.
Switched to a new branch 'codex/pro-download'
xujustinj@Justins-MacBook-Pro ExtractBench % git pull                                   
remote: Enumerating objects: 3, done.
remote: Counting objects: 100% (3/3), done.
remote: Total 3 (delta 2), reused 3 (delta 2), pack-reused 0 (from 0)
Unpacking objects: 100% (3/3), 671 bytes | 134.00 KiB/s, done.
From https://github.com/run-llama/ExtractBench
   113d93f..f492e3f  codex/pro-download -> origin/codex/pro-download
Updating 113d93f..f492e3f
Fast-forward
 README.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
xujustinj@Justins-MacBook-Pro ExtractBench % uv run extract-bench download --dataset pro
Downloading dataset from HuggingFace: llamaindex/ExtractBenchPro
Destination: data-pro
Downloading (incomplete total...): 0.00B [00:00, ?B/s]                                                                 Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
Fetching 106 files: 100%|█████████████████████████████████████████████████████████████| 106/106 [00:17<00:00,  5.91it/s]
Download complete: : 1.31GB [00:17, 53.5MB/s]              Dataset ready at: data-pro (100 test cases, revision 4ae1ffe48696)
0
Download complete: : 1.31GB [00:19, 66.4MB/s]
xujustinj@Justins-MacBook-Pro ExtractBench % uv run extract-bench status --dataset pro
Dataset: data-pro
Revision: 4ae1ffe48696

Split                     Test Cases
------------------------------------
charts_inside_documents           25
document_conventions              22
drawn_relations                   20
house_counting_rules              12
house_vocabulary                  14
long_chart_sheets                  7
------------------------------------
Total                            100
0