scripts/polygon_pdf_companion.py: parse Polygon statement PDFs into Competitive Companion

[devin-ai-integration]

Summary

scripts/polygon_pdf_companion.py (needs pip install pdfplumber): parses a Polygon/olymp.sty-generated statements PDF and POSTs each problem to Competitive Companion listeners on localhost (all default ports from CC's hosts.ts: cpbooster/Hightail/acmX/Caide/CP Editor/cph), so onsite/PDF-only contests plug into the same workflow as web parsing.

scripts/polygon_pdf_companion.py statements.pdf            # parse + POST
scripts/polygon_pdf_companion.py statements.pdf --dump     # print CC JSON
scripts/polygon_pdf_companion.py statements.pdf -p A -p C  # subset
scripts/polygon_pdf_companion.py statements.pdf --port 27121

Per problem it extracts name, TL/ML, input/output file (file-based IO becomes {"type": "file", ...}), interactivity, and sample tests, sharing one batch id per contest.

Sample extraction reads the ruled example tables via pdfplumber find_tables (with a generous intersection_tolerance — some PDFs draw rules with mismatched endpoints that otherwise split the table in half):

• one table row per test, so run-twice problems get one test per run;
• extra columns (explanation, Notes, ...) are ignored via header classification (classify_header picks input/output column indices by substring);
• interactive problems and multi-input-column communication tables (Alice's Input | Alice's Output | Bob's Input | Bob's Output) are treated as transcripts: all rows interleave into a single test;
• header-less continuation tables at a page top merge into the previous test;
• templates without in-table headers (ICPC EC-Final, UCup Finals) fall back to bordered 2-column boxes under Sample Input N/Examples headings;
• problem headers recognized: Problem A. Name, bare Problem A + title line, and A – Name (EUC).

Verified against 18 UCup/Petrozavodsk/ICPC statement PDFs (161 problems, all with non-empty samples except interactive problems with no example table) and end-to-end against a live localhost listener. Known gap: EUC-style PDFs whose samples are plain unruled text get name/limits only.

Link to Devin session: https://app.devin.ai/sessions/a55635671cc24f94a77a282da22946d1
Requested by: @ecnerwala

---

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[github-actions]

GCC Code Coverage Report

📂 Overall coverage

Metric	Coverage
Lines	🟡 8852/10410 (85.0%)
Functions	🟢 1178/1283 (91.8%)
Branches	🟡 6699/8610 (77.8%)

[beta-devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.