Meta-tests for resyntax/test failure reporting

test/private/rackunit.rkt is at 76.8% coverage (69 missed lines), and the misses are exactly the code that runs when a refactoring test fails: the code-block pretty-printing writer and the failure message rendering. The raise-syntax-error paths in test.rkt for unbound or non-statement identifiers are similarly untested. Covering these means meta-tests: deliberately-failing checks run via rackunit's run-test-case with assertions on the failure output.

This is worth more than the coverage points. Reviewers lean on this failure output when reviewing AI-written rule PRs, and right now it's the least-tested part of the test framework itself.

Part of #680.

Start with test/private/rackunit.rkt and inspect the code handling code-block pretty-printing and failure message rendering. Then review the raise-syntax-error paths in test.rkt and run the relevant checks through rackunit's run-test-case. Done means deliberately failing meta-tests assert the expected failure output and cover the unbound and non-statement identifier paths.