Unified single-call date extraction

[rajeee]

Replaces the per-page-voting + URL-shortcut date extractor with a single whole-document LLM call: the URL is passed as a hint, but the document text is the source of truth.

Fixes three failure modes in the old extractor:

• date-less URLs no longer short-circuit on a truthy {year: null} response and skip the document body
• empty pages no longer out-vote the real date in the frequency tally
• collection/upload-date contamination in page headers (e.g. energyzoning.org scrape stamps) no longer wins by repetition

Dev eval accuracy 0.43 → 0.94 (2 failing of 35).

[Copilot]

Pull request overview

This PR updates COMPASS’s ordinance date extraction to use a single whole-document LLM call (with the source URL treated only as a hint) instead of the prior per-page voting + URL-shortcut approach, and updates dev evaluation artifacts to reflect the improved accuracy.

Changes:

• Reworked DateExtractor.parse() to build one prompt from the full document text and optionally include the download URL as corroborating context.
• Replaced the old multi-response voting _parse_date logic with a simpler validation-based parser for a single response (but see review comment re: backward compatibility with existing tests/callers).
• Updated dev eval summary JSON and per-case CSV breakdown with new accuracy/cost/token numbers.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
compass/extraction/date.py	Switches to one whole-document LLM call; updates prompt/system guidance; simplifies response parsing/validation.
evals/results/dev/date_extraction_evals.json	Updates aggregate dev eval metrics reflecting the new extraction approach.
evals/results/dev/date_extraction_evals_breakdown.csv	Updates per-document dev eval outcomes and token/cost breakdown.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 61.16%. Comparing base (75cc04c) to head (5b87e43).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #466      +/-   ##
==========================================
+ Coverage   60.86%   61.16%   +0.30%     
==========================================
  Files          77       77              
  Lines        6843     6850       +7     
  Branches      670      672       +2     
==========================================
+ Hits         4165     4190      +25     
+ Misses       2561     2543      -18     
  Partials      117      117              

Flag	Coverage Δ
unittests	61.16% <100.00%> (+0.30%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.