fix: pre-existing test failures in worktrees (version-check tilde range, CodeEvaluator fixture)

[christso]

Problem

Two tests fail in worktree environments and may also be flaky on main after the v2.13.0 release:

1. checkVersion > supports tilde ranges (apps/cli)

File: apps/cli/test/version-check.test.ts:27

expect(result.satisfied).toBe(true)  // Expected: true, Received: false

The test checks checkVersion('~2.12.0') but the current version is 2.13.0, which does not satisfy a tilde range of ~2.12.0 (tilde ranges only allow patch-level changes: >=2.12.0 <2.13.0). This test broke when the version was bumped to 2.13.0.

2. CodeEvaluator > works with defineCodeJudge-based code judge (packages/core)

File: packages/core/test/evaluation/evaluators.test.ts:515

expect(result.score).toBe(1)  // Expected: 1, Received: 0

The test runs bun run test/fixtures/test-define-judge.ts which fails in worktree environments because the fixture can't resolve @agentv/eval imports — the worktree has separate node_modules that may not be fully linked.

Acceptance

• checkVersion tilde range test updated to use a range that includes the current version
• CodeEvaluator fixture test handles worktree environments (or is skipped when imports can't resolve)
• Both tests pass in both main checkout and worktree environments

Non-goals

• Changing how worktrees are set up
• Changing the version check logic itself