C++ Framework: Streaming response support

[kovtcharov]

Problem

AgentConfig::streaming exists as a field but is not implemented. Long LLM responses block for 10-30 seconds with no feedback. Applications need token-by-token streaming for responsive UIs.

Current State

• streaming field exists in AgentConfig (default: false)
• Not wired to httplib — the field is ignored
• No SSE (Server-Sent Events) parsing
• No streaming callback

Proposed Solution

SSE parser

Parse text/event-stream responses from OpenAI-compatible endpoints. Extract data: lines, handle [DONE] sentinel.

Streaming HTTP

Use httplib's set_response_handler() with chunked transfer encoding. Feed chunks to SSE parser.

Token callback

using StreamCallback = std::function<void(const std::string& token)>;

Called for each token as it arrives during LLM inference.

OutputHandler integration

New virtual method:

virtual void printStreamToken(const std::string& token);

TerminalConsole prints immediately; SilentConsole buffers silently.

Response accumulation

Collect all tokens into a complete response string for JSON parsing after the stream completes. The existing multi-strategy JSON parser operates on the complete response.