test(smollm3): add unit tests for HumanEval+ and MBPP+ evaluation utilities (#137)

[SiddardhaShayini]

Fixes #137

What changes are proposed in this PR?

• Adds comprehensive unit tests for HumanEval+ and MBPP+ evaluation utilities under text/evaluation/smollm3/.
• Covers critical edge cases including empty/whitespace model outputs, markdown code block variations, unclosed code fences, and malformed EvalPlus JSON result payloads.
• Runs entirely on CPU without requiring GPU acceleration or heavy model/dataset downloads.

How to test?

You can run the tests locally via pytest:

cd text/evaluation/smollm3
pytest test_eval_plus.py -v