feat(scoring): add ergonomics helpers for TemplateMetric authoring

[wprazuch]

Completes the ~20-30 LOC authoring promise by giving TemplateMetric subclasses ready-made helpers. Stacked on wprazuch/metric-abstractions.

New:

• TemplateMetric._render(template, input) — Jinja2 rendering with both NEL-native ({{ response }}, {{ target }}, {{ metadata.* }}) and SDK-native ({{ output_text }}, {{ reference }}) variable names, so templates authored against either vocabulary work unchanged. Strict undefined-variable handling raises instead of silently rendering empty.

• CorpusTemplateMetric(TemplateMetric) — base class for metrics with both row-level and corpus-level scores. Subclasses implement _score() and _corpus_score(); defaults wrap each in a MetricResult. score_names() includes both '' and '_corpus'. Empty inputs -> None.

• SecretsMixin — mixin that satisfies MetricWithSecrets protocol. Subclasses declare secret_env_vars: ClassVar[tuple[str, ...]]. Secrets are eagerly loaded from os.environ at construction, with async resolve_secrets() as a fallback path (NMP Platform flow). Resolved values are stored as SecretStr private attrs; get_secret(env_var) returns the plaintext value or None.

Tests (+18 new, 42 total):

• _render: NEL-native names, SDK-native aliases, metadata/config access, StrictUndefined raises on missing variables.
• CorpusTemplateMetric: satisfies both Metric + CorpusMetric protocols, row-level default, corpus-level default, empty-inputs, score_names includes both.
• SecretsMixin: satisfies MetricWithSecrets, declares env vars, reads env at construction, async resolver fills gaps, resolver is skipped when already loaded.
• Target ergonomics proof: _TinyLengthMetric in ~15 LOC of user code demonstrates the authoring pattern.

NMP can pick these up as they land. The contract API is stable — helpers are additive (subclassing-based), so SDK concrete metrics can adopt them incrementally without breakage.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: e25a603c-6b7c-4a79-88f5-ecf99466765e

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch wprazuch/metric-abstractions-ergonomics

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}