Split test cases by failure/success category

[leo-aa88]

Summary

Restructure the test corpus so language behavior is grouped by execution stage instead of keeping everything in one flat bucket.

Why

Right now tests mostly validate output or error text, which is already useful. But as the language grows, we should be explicit about whether a program is expected to:

• run successfully
• fail in parsing
• fail in semantic analysis
• fail at runtime

That makes the language behavior clearer and prevents vague “some error happened” coverage.

Proposed changes

• Split test cases into categories:

  • success/
  • syntax_error/
  • semantic_error/
  • runtime_error/

• Update the test harness to:

  • assert expected exit behavior
  • assert stdout/stderr depending on category
  • make stage-specific expectations clearer

Acceptance criteria

• Test cases are grouped by category
• Harness supports category-aware assertions
• Existing tests are migrated without behavior loss
• Adding new tests in each category is straightforward

Notes

This is mainly a test-organization/specification issue, not a language-feature issue.