feat: Add completion attribute to ContentFilterFinishReasonError for parity with LengthFinishReasonError

[PupiBott]

Problem

When finish_reason == "content_filter", the SDK raises ContentFilterFinishReasonError without providing access to the underlying ChatCompletion object. This makes it impossible to:

• Track usage/costs via completion.usage
• Log completion ID for debugging
• Access any response metadata

This is inconsistent with LengthFinishReasonError, which includes completion as a public attribute.

Solution

Modified ContentFilterFinishReasonError to accept and expose completion parameter, matching the pattern established by LengthFinishReasonError.

Changes:

1. Updated ContentFilterFinishReasonError constructor to accept completion parameter
2. Updated parsing call site to pass completion
3. Updated streaming call site to pass completion_snapshot
4. Added comprehensive tests

Backward Compatibility

Existing try/except blocks will continue to work. The only breaking change would be code that manually instantiates ContentFilterFinishReasonError() without parameters, which is not a realistic use case.

Related Issues

• Fixes Include response information in ContentFilterFinishReasonError #2690

Example Usage

try:
    result = await client.beta.chat.completions.parse(...)
except ContentFilterFinishReasonError as e:
    # Now you can access:
    print(f"Completion ID: {e.completion.id}")
    print(f"Tokens used: {e.completion.usage.total_tokens}")
    print(f"Model: {e.completion.model}")

Testing

Added two new tests:

• test_parse_content_filter_includes_completion: Verifies completion access and usage info
• test_parse_content_filter_backward_compatibility: Ensures existing code still works

All tests pass including existing LengthFinishReasonError tests.