Plugin system for custom metrics via entry points

[rudrakshkarpe]

Summary

Allow users to register custom metrics without forking the package, using Python entry points or a simple registry API.

Scope

• dual_llm_bench.metrics entry point group for third-party metrics
• register_metric() API for programmatic registration
• Discovery in built_in_metrics() + registered/entry-point metrics
• CLI --metrics flag to select specific metrics by name

Acceptance Criteria

• Entry point discovery works for installed packages
• register_metric() works for inline usage
• CLI supports metric selection
• Documentation with example custom metric