Unified networking layer: rate limiting, retries, error handling

[JordanCoin]

Every adapter creates its own httpx.AsyncClient with ad-hoc error handling. Need a shared networking layer.

Problems

• No unified rate limiting — each adapter guesses independently
• 404s not handled gracefully (ProPublica blows up on special chars)
• No retry with backoff on transient failures
• No connection pooling across adapters
• DEMO_KEY shared across all api.data.gov users — we hit limits fast
• Crossref runs all entities even when dedup/confidence filters should reduce them

Solution

• Shared APIClient class wrapping httpx with per-host rate limiters
• Token bucket or leaky bucket rate limiting per domain
• Automatic retry on 429/503 with exponential backoff
• URL-encode entity names properly before sending
• Connection pooling via shared client instances
• Log warnings (not errors) for expected failures