Add thread-safety support (connection pooling)

[CaptainDriftwood]

Description

IcapClient is not thread-safe. Multiple threads sharing a client instance could corrupt socket state.

Current Issues

• Socket interleaving if two threads call methods concurrently
• Check-then-act races in connection logic
• State corruption during disconnect

Implementation Options

1. Lock-based (threading.RLock) - Simple, serializes operations
2. Thread-local storage - Each thread gets own client
3. Connection pool (IcapClientPool) - Bounded resources, efficient reuse
4. Stateless - New socket per operation

Recommendation

Implement Option 1 (lock-based) as default for basic thread-safety, and provide Option 3 (connection pool) as IcapClientPool for high-throughput use cases.

Python Compatibility

All approaches work across Python 3.8-3.14. Note: Python 3.13+ free-threaded mode (PEP 703) makes explicit locking more critical.

Priority

Future enhancement - Not blocking release.