fix(reward): add missing detail page patterns for 4 plugins

[bittoby]

Problem

DETAIL_PAGE_PATTERNS in core/reward.py only covers the 4 original plugins (CoinGecko, Stooq, Taostats, Weather). The 4 newer plugins — ArXiv (#9), OpenLibrary (#2), Open-Meteo (#5), and HackerNews — have no detail page patterns at all.

This means:

• is_detail_page() always returns False for these plugins
• The DETAIL_PAGE_VISIT reward signal (+0.03) never fires
• _extract_asset_from_url() returns None, so no asset confirmation tracking occurs
• _normalize_url() strips query params that identify resources on HackerNews (?id=) and Open-Meteo (?latitude=&longitude=), causing false "repeated URL" penalties

Impact: Agents browsing detail pages on 4 out of 8 active plugins receive no exploration reward for drilling into content. This silently degrades RL training quality for half the plugin portfolio.

Summary

• Added detail page regex patterns to DETAIL_PAGE_PATTERNS for all 4 missing plugins
• Added asset ID extraction to _extract_asset_from_url() for the same plugins
• Added query param preservation in _normalize_url() for HackerNews and Open-Meteo

What changed

Plugin	Detail page pattern	Asset ID example
ArXiv	/abs/2603.16870	2603.16870
OpenLibrary	/works/OL103123W/Dune	ol103123w
Open-Meteo	/en/docs?latitude=35.68&longitude=139.65	35.68,139.65
HackerNews	/item?id=12345	12345

Each pattern was verified against the actual plugin URL parsing logic (arxiv.py, openlibrary.py, openmeteo.py, hackernews.py) and live website URL structures.

Test plan

• 30 existing reward tests pass, zero regressions
• Manual verification: is_detail_page() returns True for detail URLs, False for list/search/homepage URLs across all 4 plugins
• _extract_asset_from_url() correctly extracts asset IDs from all new URL formats
• _normalize_url() preserves query params that identify resources (HN id, Open-Meteo latitude/longitude)

[bittoby]

@angosr please review this PR. thanks

[bittoby]

Hi, @angosr any updates for me, please. It took long time since I submitted this PR.

[angosr]

Review: PR #17 — APPROVE

Significance Gate: PASS

Fixes a real bug: 4 of 8 active plugins had no detail page patterns, meaning agents received zero exploration reward for visiting detail pages on ArXiv, OpenLibrary, Open-Meteo, and HackerNews. This silently degrades RL training quality.

Independent verification (not relying on PR's own test claims)

1. Regex patterns — 10/10 correct (tested independently)

URL	Expected	Result
arxiv.org/abs/2603.16870	detail=True	✅
arxiv.org/abs/2604.00005 (5-digit)	detail=True	✅
arxiv.org/abs/2603.16870v2 (version suffix)	detail=True	✅
arxiv.org/pdf/2603.16870 (PDF, not abs)	detail=False	✅
arxiv.org/list/cs.AI/new (listing)	detail=False	✅
openlibrary.org/works/OL103123W/Dune	detail=True	✅
openlibrary.org/search?q=dune	detail=False	✅
open-meteo.com/en/docs?latitude=-33.87&longitude=151.21 (negative coords)	detail=True	✅
open-meteo.com/en/docs (no params)	detail=False	✅
news.ycombinator.com/item?id=12345&goto=news (extra params)	detail=True	✅

2. URL normalization — preserves essential params, strips noise

• HN ?id=12345&p=2 → ?id=12345 ✅
• OpenMeteo ?latitude=35.68&longitude=139.65&hourly=temp → ?latitude=35.68&longitude=139.65 ✅
• OpenMeteo missing longitude → falls through to default (strips all params) ✅

3. Asset extraction — correct for all edge cases

• ArXiv v2 suffix → extracts 2603.16870 (strips version) ✅
• OpenMeteo negative coords → extracts -33.87,151.21 ✅

4. Existing tests: 30/30 passed, zero regressions

Minor note (non-blocking)

No new automated tests added for the 4 new plugins' patterns. The existing 30 tests only cover the original 4 plugins. Recommend adding parametrized tests in a follow-up, but the patterns are simple regex additions following the established pattern and verified above.

[bittoby]

@angosr Thanks for approving. would you merge this PR?