## Why

Screenshot OCR inherits its endpoint, model, and key from the summary agent, but it bypasses the shared Responses client and ignores `Agent:UseStreaming`. With streaming enabled, the screenshot client still expects one JSON document and cannot consume the configured LiteLLM Responses event stream.

## What Changes

- Make screenshot OCR inherit the summary agent's streaming transport selection.
- Send screenshot image input through the same OpenAI Responses SDK and Agent Framework adapter used by the summary client.
- Preserve the existing non-streaming screenshot OCR path when streaming is disabled.
- Add behavior coverage for streamed and non-streamed screenshot OCR responses.

## Capabilities

### New Capabilities

- None.

### Modified Capabilities

- `meeting-summary`: Require screenshot OCR to honor the configured agent streaming transport while preserving image input and OCR metadata parsing.

## Impact

The change affects the screenshot OCR client, the shared LiteLLM Responses message translation, focused tests, and agent configuration documentation. It does not change the local HTTP API or screenshot note format.