fix: return ExactMatch mismatches for non-JSON primitives

[Andrew Barnes (Bortlesboat)]

Summary

ExactMatch()(output=None, expected={"amount": 100}) currently raises TypeError instead of returning a mismatch score. The same happens for booleans and numbers compared with a dict or list, including when the arguments are reversed. These comparisons now return 0, using the existing normalization fallback.

Validation

• All 8 new parameterized cases failed before the fix.
• pytest py/autoevals/test_values.py py/autoevals/test_json.py -q: 16 passed after the fix.
• Black, Flake8, and git diff --check passed for the changed files.
• Manually checked async evaluation, partial scorers, and existing bytes/bytearray JSON normalization.

The broader Python suite was attempted but is not green in this Windows environment: optional LiteLLM is missing, live API tests lack valid credentials, and the checkout's template symlink is not resolved.

Post-deploy validation

No service deployment is involved. After a package release, the example above should return a score of 0; I can help investigate any reported normalization regression.