fix: resolve ZeroDivisionError and NameError in OpenAICloseSetClsEvaluator.print_results

[octo-patch]

Problem

OpenAICloseSetClsEvaluator.print_results() in pointllm/eval/evaluator.py contains two bugs that cause crashes when total_predictions == 0 (e.g. evaluation on an empty result set):

1. ZeroDivisionError: A stray accuracy = self.correct_predictions / self.total_predictions * 100 line was placed outside the if/else guard block (line 568 before this fix). This overwrites the accuracy = 0 safety value set in the if branch, then immediately raises ZeroDivisionError when total_predictions == 0.

2. NameError: clean_accuracy was only assigned inside the else branch. When the if branch executes (i.e. total_predictions - invalid_responses == 0), clean_accuracy was never defined, causing NameError: name 'clean_accuracy' is not defined on the subsequent print call.

Note: the companion method save_results() already handles both variables correctly in both branches — this fix brings print_results() into alignment with that implementation.

Solution

• Remove the duplicated accuracy assignment outside the if/else block.
• Initialize clean_accuracy = 0 in the if branch, matching the pattern already used in save_results().

Additional fixes

• Fix two typos: "unale to parse" → "unable to parse" (in OpenAICloseSetClsEvaluator.parse_gpt_response_evaluate and OpenAIObjectCaptioningEvaluator.parse_gpt_response_evaluate).

Testing

Manually verified both bugs with a minimal reproducer:

# Before fix: raises ZeroDivisionError when total_predictions == 0
# Before fix: raises NameError for clean_accuracy when if-branch executes
# After fix: returns accuracy=0, clean_accuracy=0 safely