Add injectable client factories for recursive RLM runtimes

[vishaltandale00]

Summary

• add an optional client_factory seam for custom BaseLM runtimes
• propagate the factory through recursive children and max-depth fallback
• route unnamed recursive calls to the configured child backend
• document custom runtime injection

Why

Applications that already own an agent runtime need to reuse authentication, tools, model routing, and tracing without modifying rlms' global backend registry. The factory must remain active at every recursion depth.

Tests

• 51 focused recursion tests passed, 1 skipped
• 276 non-Docker tests passed, 8 skipped
• Docker tests were excluded because the local Docker fixture could not mount /workspace/_exec.py