[psincraian]

Fixed dependency conflict where library A could not use Klyne if its dependency (library B) already used Klyne. The SDK now supports multiple concurrent clients in the same Python process.

Changes

Architecture:

• Replaced single global _client with _clients dictionary (keyed by project name)
• Added _default_client for backward compatibility with module-level API
• Each project gets its own isolated KlyneClient instance

API Enhancements:

• init() now returns KlyneClient instance for direct use
• Module-level functions (track, flush, etc.) still work for backward compatibility
• Libraries can use either module-level API or instance-based API

Auto-initialization:

• Removed automatic self-analytics initialization on import
• Self-analytics now opt-in via KLYNE_SELF_ANALYTICS=1 environment variable
• Prevents unwanted side effects and background threads on import

Testing

• All existing integration tests pass (backward compatibility verified)
• Added comprehensive test suite in test_multiple_clients.py
• Tested nested library scenario (Library A depends on Library B)

Breaking Changes

None - fully backward compatible. Existing code continues to work unchanged.

Migration Guide

For libraries that want isolation from other Klyne users:

# Old style (still works)
import klyne
klyne.init(api_key="...", project="my-lib")
klyne.track("event")

# New style (recommended for libraries)
import klyne
client = klyne.init(api_key="...", project="my-lib")
client.track("event")

Fixes issue where only one library per process could use Klyne analytics.