Release 0.2.0 with schema-enforced judge responses

[dpaluy]

Summary

• Bumps RubricLLM::VERSION from 0.1.2 to 0.2.0.
• Adds 0.2.0 changelog notes for schema-backed judge responses and explicit judge contract failures.
• Applies RubyLLM structured output schema enforcement when available.
• Raises RubricLLM::JudgeError for malformed, missing-score, non-numeric, or out-of-range judge responses.
• Adds offline regression coverage for schema usage and judge response contract failures.

Why

Judge scoring should be governed by an API/schema contract, not prompt compliance alone. This release makes invalid judge responses fail explicitly instead of becoming nil scores or clamped values.

Changes

release: surface judge contract failures in 0.2.0

• What: Adds a shared metric response schema, applies it through chat.with_schema, validates parsed judge responses, updates tests, bumps the gem version, and documents the release in CHANGELOG.md.
• Why: GitHub issue Schema-enforce judge responses #3 requires schema-backed judge responses and explicit failures for invalid judge contracts.

Testing

• bundle exec rake test
• bundle exec rubocop
• bundle exec rake
• gem build rubric_llm.gemspec
• git diff --check

Risk & Rollout

• Risk: Medium - this intentionally changes invalid judge outputs from nil/clamped results into RubricLLM::JudgeError failures.
• Rollback: Revert e404b72 to restore the previous parsing and version/changelog state.
• Monitoring: Watch downstream evaluation/test integrations for newly surfaced JudgeError failures from providers that return malformed or out-of-range payloads.

Closes #3