Add streaming prompt support for GPT providers (OpenAI, Gemini, Ollama, Anthropic)

[nmaguiar]

Motivation

• Add opt-in streaming responses to the genai implementations so callers can receive progressive output (deltas) from providers that support SSE/streaming.
• Reuse existing $rest(...).get2Stream/post2Stream plumbing and local stream helpers to parse SSE/NDJSON payloads and integrate with current conversation/tool-calling flow.

Description

• Implemented an SSE/NDJSON helper _readSseStream and provider-specific _requestStream functions that call $rest(...).get2Stream/post2Stream with Accept: text/event-stream where appropriate.
• Added rawPromptStream and promptStream methods to each provider implementation (OpenAI, Gemini, Ollama, Anthropic) to parse streamed payloads, aggregate content and events, surface tool-call deltas, and capture stats.
• Exposed streaming APIs on the public interfaces by adding ow.ai.gpt.prototype.promptStream, ow.ai.gpt.prototype.rawPromptStream, and $gpt.promptStream / $gpt.rawPromptStream wrappers.
• Kept non-streaming behavior unchanged and added odoc-compatible entries for the new streaming methods.

Testing

• No automated tests were executed as part of this change (no CI/test run performed).
• To validate manually with automation, create a streaming-capable provider client and call rawPromptStream/promptStream (for example via $gpt({...}).promptStream(...)), or run the repository test suite with cd tests && ojob autoTestAll.yaml as recommended by project guidelines.