fix(OpenAI): handle undocumented text web search results- #10

[e4se]

What:

• Bug Fix
• New Feature

Description:

Hotfixes a regression in Responses API web search result parsing.

web_search_call.results can contain undocumented text_result entries in addition to the documented image_result entries. Treating every result as an image attempted to read image_url from text results and caused an Undefined array key "image_url" exception.

• Dispatch web_search_call.results entries by their type discriminator.
• Parse image_result entries into typed OutputWebSearchToolCallImageResult objects.
• Parse text_result entries into typed OutputWebSearchToolCallTextResult objects.
• Preserve optional text result titles, URLs, and snippets.
• Add regression coverage for both image and text result payloads.

The OpenAI documentation currently describes image result payloads but does not document the text_result variant: Image search results.

Tests:

• Unit test suite: 1,140 passed (5,146 assertions).
• Targeted web search tests: 10 passed (67 assertions).
• PHPStan, code style, and PHP syntax checks passed.

[iBotPeaches]

Slight breaking change here if people depended on the old class, but not much we can do if it split out to text vs image.

[e4se]

Yes, my bad — I didn’t catch that this broke the default web search without image(