Multimodal Inference API Expansion

Natively embed audio, video, and PDF content through the inference API.

**What the feature is**
An expansion of the Elastic inference API spec to support audio, video, and PDF embedding types, alongside updated OpenAI integration specs and improved default timeouts for chat and completion tasks.

**Who is this feature for**
Developers and enterprise software engineers building AI applications on non-text data.

**Expected outcome**
Handling audio, video, and PDF data today usually means juggling multiple custom tools and workflows. Native support removes that complexity, so teams can build multimodal AI applications faster without stitching together separate pipelines.

**Key user stories / use cases**
Building a search engine that finds specific moments inside video tutorials, scanning audio call recordings for compliance, and searching across large libraries of corporate PDF reports.