fix(batch): accept empty structured results

[hsusul]

Summary

Allow valid empty structured objects in batch results to reach Pydantic validation.

Problem

BatchProcessor.parse_results() treats an extracted {} response as an
extraction failure. Models whose fields have defaults therefore receive a
BatchError instead of a successfully validated result.

This affects both OpenAI JSON-schema batch responses and Anthropic tool-use
batch responses.

Root cause

The parser used a truthiness check for extracted_data, conflating an empty
dictionary with the None sentinel returned when no structured result could
be extracted.

Fix

Check explicitly for None. Empty dictionaries are now validated normally,
while absent and malformed results retain the existing error behavior.

Test plan

• Added a regression covering an OpenAI batch response containing {}
• Added the equivalent Anthropic tool-use response
• Verified both fail before the fix and pass afterward
• Ran the focused batch suite, broad offline suite, Ruff, ty, and pre-commit

Compatibility considerations

This is backwards-compatible. Only valid empty mappings change behavior;
missing or malformed results continue to produce BatchError.

Issue ticket number and link

N/A — independently discovered; no matching issue found

[jxnl]

Consolidated and shipped in #2495. Closing this focused patch as superseded; thank you for the contribution.