Expose RLM metrics to verifiers

[snimu]

Description

Previously, ComposableEnv metrics were namespaces but never added to state["metrics"] and thus never made visible. This PR fixes that, especially with an eye on the RLM harness.

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

Medium Risk
Touches scoring/aggregation paths by adding a new cleanup rubric and changing RubricGroup to treat None rewards as 0, which could affect how rewards/metrics are combined across rubrics.

Overview
ComposableEnv now exposes harness-collected numeric metrics via state["metrics"]. It collects prefixed numeric values into state["_harness_metrics"] during post_rollout, then a new HarnessMetricsRubric merges them into state["metrics"] during rubric cleanup when metrics_path is configured.

Rubric aggregation is made more robust by treating None reward values as 0.0 in RubricGroup for both score_rollout and score_group, avoiding crashes when a rubric only contributes metrics.

Tests are updated/added to assert RLM harness metrics end up in state["metrics"] after cleanup and that scoring still succeeds with the new metrics rubric.

^{Reviewed by Cursor Bugbot for commit a3a4f71. Bugbot is set up for automated code reviews on this repo. Configure here.}

[mikasenghaas]

i dont like private classes

[snimu]

Made public

[mikasenghaas]

mmh doing this as cleanup func seems like an anti-pattern. can we not just read harness metrics into metrics directly and not have a HarnessMetricRubric at all?

[snimu]

Discussed internally: this is the only way to do this without touching core verifiers functionality, which we should be very careful about.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 7e2973c. Configure here.}