Avoid duplicate guardrail metric history

[sylvesterkaczmarek]

Summary

• Stop appending the current metric snapshot a second time inside ThresholdAdjuster.calculate_adjustment().
• Keep threshold history, adjustment history, and explicit update_metrics() behavior unchanged.

Motivation

The feedback loop already records every evaluation result through ThresholdAdjuster.update_metrics(): once after the initial eval and once after each subsequent eval. calculate_adjustment() then appends that same GuardrailMetrics object again before changing the threshold.

As a result, metrics_history contains duplicate entries for every metric snapshot that triggers an adjustment. That makes the tuning history inaccurate for diagnostics or future consumers even though the current report's first/last metric lookup happens to mask the duplication.

There is already one explicit method responsible for recording metric observations, so adjustment calculation should not record the same observation again.

Validation

• initial eval metric -> recorded once by update_metrics()
• each re-eval metric -> recorded once by update_metrics()
• threshold/adjustment history -> unchanged
• convergence and threshold calculation -> unchanged

Self-review

• One-line deletion in one guardrail-tuner helper.
• No API, dependency, notebook, registry, or configuration changes.
• Searched open PRs for an existing metric-history duplication fix and found none.

Maintainers may modify the branch if needed.