Expand /evaluate response definition: formatted content and case matching

[m-messer]

Summary

• Adds a reusable FormattedContent schema (paths/shared/schemas/FormattedContent.yml) representing a submission's normalized value in multiple, service-defined formats keyed by an arbitrary format id (e.g. latex/sympy for MATH, formatted/ast for CODE) — resolves the response-format ambiguity raised in Response Definition #31 (Lambda Feedback's responseSympy/responseLatex pattern) without hardcoding any specific format names.
• Wires it into Feedback.response, and adds Feedback.matchedCase (index into a service-defined case-matching list, e.g. Lambda Feedback's cases config) — both grounded in real example responses from a Lambda-Feedback-compatible service (compareExpressions).
• Adds a mathExampleResponse example to evaluate.yml demonstrating both new fields in matched and unmatched states.
• Doc-only clarification on PreSubmissionFeedback.yml / EvaluateRequest.yml pointing out that pre-submission preview content comes back via Feedback.response, not the request-side toggle schema.
• Fixes a pre-existing stray-character typo in EvaluateHealthResponse.yml that was breaking YAML parsing and blocking lint.

Explicitly out of scope (by design): no formal Cases/Case schema — case-based matching stays a Lambda-Feedback-specific, informal convention under the free-form configuration object.

Closes #31.

Test plan

• npm run lint passes (0 errors; only pre-existing unrelated warnings remain)
• npm run bundle succeeds; confirmed FormattedContent and the new fields resolve correctly in dist/openapi.yml

🤖 Generated with Claude Code

https://claude.ai/code/session_017yjEAviVJipdWqJ9RgjnSm