feat(stream): add comprehensive streaming module parity with npcpy.streaming

[celeria-ai]

Summary

Adds a complete streaming module to npcts that mirrors npcpy's streaming.py for feature parity.

What's Added

Core Streaming Infrastructure

• SSE formatting: formatSSEEvent(), formatSSERaw() with 4096-byte padding for Eventlet/gevent flush compatibility
• Chunk parsing: parseStreamChunk() handles OpenAI, Ollama, llama.cpp, and plain dict formats
• Message cleaning: cleanMessagesForLLM() fixes orphaned tool_calls/tool results that cause API errors
• System prompt management: ensureSystemPrompt() manages system messages with tool capability awareness

Tool Support

• Tool resolution: resolveNPCTools() resolves jinx catalog, MCP clients, and Python functions
• Tool execution: executeTool() runs jinx, MCP, and Python tools with proper error handling
• Tool call accumulation: accumulateToolCallDeltas() merges streaming tool_call deltas

Streaming Generators

• createChatStream(): Basic streaming without tool execution
• createToolAgentStream(): Agentic streaming with tool-calling loop (mutates messages in place)
• createJinxStream(): Jinx-based agentic streaming loop with checkLLMCommand integration

Type Definitions

Full TypeScript interfaces matching Python types: StreamEvent, ChatMessage, ToolCall, ToolCallDelta, ToolExecutor, ToolSchema, SSEChunk, StreamConfig, ParsedChunk, etc.

Backward Compatibility

Snake_case aliases for all camelCase exports matching npcpy naming conventions.

Line Count

~650 lines of TypeScript implementing npcpy.streaming.py (~842 lines of Python).

References

• Mirrors: https://github.com/NPC-Worldwide/npcpy/blob/main/npcpy/streaming.py

[celeria-ai]

Documentation Gap Assessment

This PR introduces a comprehensive streaming module (~350 lines) but no documentation exists for the new functionality.

Identified Gaps

1. README.md - No streaming API documentation for 6 new public exports
2. No examples/ - No usage examples for SSE or plain streaming backends
3. No parity mapping - Developers familiar with npcpy.streaming have no reference

Code Quality Notes

• ⚠️ formatMessageForLLM() has incomplete JSDoc (ends with //)
• ⚠️ 350+ lines of new code with no test coverage
• ✅ Build passes cleanly
• ✅ Implementation matches npcpy.streaming parity as claimed

Recommendation

Please add before merge:

1. README section documenting the streaming API with examples
2. JSDoc improvements (fix incomplete @returns tag)
3. Tests for the streaming module

Verdict: Code looks correct but docs gap is significant for a "parity" PR that developers will reference.

Full details logged to agent_memory/npc_projects/daily_log/2026-06-10.md.