Fix flaky defineCodeJudge test failing in full suite

[christso]

Problem

The test CodeEvaluator > works with defineCodeJudge-based code judge in packages/core/test/evaluation/code-evaluator-file-backed.test.ts passes when run individually but fails when run as part of the full test suite.

This is a test isolation issue — some shared state from other tests affects this test's outcome.

Reproduction

# Passes individually:
bun test packages/core/test/evaluation/code-evaluator-file-backed.test.ts

# Fails in full suite:
bun test

Expected Behavior

The test should pass reliably regardless of execution order or which other tests run alongside it.

Likely Cause

Test isolation issue — possibly shared module state, global mocks, or file system side effects from other tests leaking into this one.

Acceptance Signals

• Test passes consistently in full suite (bun test) across multiple runs
• No other tests broken by the fix
• Root cause documented in commit message

Related

Discovered during #431 / PR #436 implementation (855/856 tests passing, this was the 1 pre-existing failure).

[christso]

Fixed by #450 — the root cause was the same as #449: the test-define-judge.ts fixture imported from @agentv/eval via workspace resolution, which was unreliable when other tests ran in the same suite (shared module state affecting package resolution). The fix changed the import to a relative path (../../../eval/src/index.ts), making it independent of workspace symlinks. Verified: 875 core tests pass consistently across multiple full-suite runs with 0 failures.