Add immediate sub-LLM trajectory steps and incremental metrics

[snimu]

Description

Instead of adding the sub-LLM TrajectorySteps and the metric during cleanup, this PR changes that to happen incrementally as the sub-LLM responses come in; and the metrics are also updated incrementally as the main model's steps come in.

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Test improvement

Testing

• All existing tests pass when running uv run pytest locally.
• New tests have been added to cover the changes

Checklist

• My code follows the style guidelines of this project as outlined in AGENTS.md
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

Additional Notes

---

Note

Implements immediate insertion of sub-LLM turns and incremental metric tracking throughout RLM execution.

• Adds update_rlm_metrics_from_step, _extract_tokens_from_response, and _ensure_rlm_metric_state to track sub-LLM/main turns, tokens, tool calls, batches, and REPL timing incrementally; overrides add_trajectory_step to update metrics on each step
• _handle_sub_llm_request now constructs per-turn TrajectorySteps with extras and either immediately calls add_trajectory_step (when enabled) or updates metrics only; active_rollouts stores state for this
• Tracks REPL execution timings via _update_rlm_repl_metrics; initializes metric fields in setup_state
• Context warning in call_python_repl now uses the last main-model step (ignores sub-LLM steps)
• Simplifies cleanup_rlm_state to just remove rollout and release tunnels (metrics/steps no longer computed here)
• Tests updated/added for immediate sub-LLM trajectory insertion, disabled behavior, token/tool-turn accumulation, and main-step-only context warning

^{Written by Cursor Bugbot for commit d6beb5c. This will update automatically on new commits. Configure here.}