feat(enrichers): Arabic media enrichers (Sabq, Argaam, Al Arabiya, Nitter)

[SocialMDev]

Summary

Adds 8 enrichers that surface Arabic-language mentions of an Individual or a Phrase (topic) and link them into the graph as Website nodes with source-specific relationship labels.

Enricher	Input → Output	Relationship	Source
individual_to_sabq / phrase_to_sabq	Individual / Phrase → Website	MENTIONED_IN_SABQ	sabq.org HTML search
individual_to_argaam / phrase_to_argaam	Individual / Phrase → Website	MENTIONED_IN_ARGAAM	argaam.com HTML search
individual_to_alarabiya / phrase_to_alarabiya	Individual / Phrase → Website	MENTIONED_IN_ALARABIYA	Google News RSS (site:alarabiya.net)
individual_to_arabic_tweets / phrase_to_arabic_tweets	Individual / Phrase → Website	MENTIONED_ON_TWITTER_AR	Nitter mirrors → Google dork fallback

What changed

flowsint-enrichers/
├── pyproject.toml                    +1 dep: defusedxml
├── src/
│   ├── flowsint_enrichers/
│   │   ├── individual/
│   │   │   ├── to_sabq.py            NEW
│   │   │   ├── to_argaam.py          NEW
│   │   │   ├── to_alarabiya.py       NEW
│   │   │   └── to_arabic_tweets.py   NEW
│   │   └── phrase/                   NEW dir (Phrase-input variants)
│   │       ├── __init__.py
│   │       ├── to_sabq.py
│   │       ├── to_argaam.py
│   │       ├── to_alarabiya.py
│   │       └── to_arabic_tweets.py
│   └── tools/arabic_media/           NEW dir
│       ├── __init__.py
│       ├── sabq.py                   SabqTool
│       ├── argaam.py                 ArgaamTool
│       ├── alarabiya.py              AlArabiyaTool (uses defusedxml for RSS)
│       └── nitter.py                 NitterArabicTool
└── tests/enrichers/
    ├── test_arabic_sabq.py
    ├── test_arabic_argaam.py
    ├── test_arabic_alarabiya.py
    └── test_arabic_tweets.py

Why a new phrase/ category

Sabq / Argaam / Al Arabiya all support searching for topics, not just people. Phrase was already in flowsint-types but had no enrichers — this PR adds the first set. Topic search is useful for journalists / OSINT investigators tracking issues rather than individuals.

Security notes

• defusedxml is used in AlArabiyaTool for parsing Google News RSS, to avoid XXE / billion-laughs attacks on untrusted XML. Added as a dependency (>=0.7,<0.8).
• All four scrapers respect a 10s timeout and degrade gracefully on non-200 responses.
• Nitter tool tries each mirror in NITTER_INSTANCES then falls back to a Google dork; tests cover both paths via mocking.

Demo

Brought up docker-compose.dev.yml infra (postgres + redis + neo4j) and ran individual_to_sabq against real Neo4j with HTTP mocked to return 3 fixture article hits for "Faisal Aldeghaither":

MATCH (i:individual)-[r:MENTIONED_IN_SABQ]->(w:website)
RETURN i.`nodeProperties.full_name` AS person, type(r) AS rel, w.`nodeProperties.url` AS url, w.`nodeProperties.title` AS title;

person, rel, url, title
"Faisal Aldeghaither", "MENTIONED_IN_SABQ", "https://sabq.org/news/202605/saudi-vision-update-1", "تحديثات رؤية المملكة 2030"
"Faisal Aldeghaither", "MENTIONED_IN_SABQ", "https://sabq.org/news/202605/saudi-vision-update-2", "مقابلة حصرية"
"Faisal Aldeghaither", "MENTIONED_IN_SABQ", "https://sabq.org/news/202605/saudi-vision-update-3", "تقرير اقتصادي"

The Neo4j Browser visualisation showing the Individual → 3 Website subgraph with MENTIONED_IN_SABQ edges and Arabic article titles is attached in the first PR comment.

Test plan

• pytest tests/enrichers/test_arabic_*.py — 19 new tests, all green
• Existing tests/enrichers/test_registry.py still passes (21/21 with new suite)
• All 8 enrichers register via @flowsint_enricher and appear in ENRICHER_REGISTRY
• Postprocess writes to real Neo4j (verified with cypher-shell on demo stack)
• Dedup logic: re-running with duplicate URLs creates each Website + relationship only once
• No live network calls in tests — SabqTool, ArgaamTool, AlArabiyaTool, NitterArabicTool are mocked

Notes for maintainer

• The existing to_domains.py reference enricher served as the architectural template (preprocess / scan / postprocess split, @flowsint_enricher decorator, module-level InputType / OutputType re-export). I tried to match style and structure exactly; happy to adjust if you want different conventions for the new phrase/ category.
• HTML selectors for sabq.org and argaam.com are defensive (multiple fallbacks via comma-separated CSS selectors) but will need maintenance if those sites change their markup.
• Discord-friendly: I can open follow-ups to add similar enrichers for other Arabic media (Asharq, Okaz, Riyadh Daily) if the pattern lands well.

[SocialMDev]

Demo evidence — individual_to_sabq executed against the dev docker-compose.dev.yml infra (postgres + redis + neo4j), HTTP mocked with 3 fixture article hits. Neo4j Browser shows the resulting subgraph with the Faisal Individual node connected via MENTIONED_IN_SABQ to 3 Arabic-titled Website nodes (screenshot attached separately by branch owner).