feat(mcp): Phase 3 — MCP resources and prompts

[staging-devin-ai-integration]

Summary

Follow-up from #358 (Phase 1 landed in #359, Phase 2 in #360).

Phase 3 exposes pipeline state as MCP resources and adds prompt templates.

MCP Resources

• streamkit://sessions/{id}/pipeline — live pipeline YAML (subscribable)
• streamkit://sessions/{id}/metrics — per-node throughput/latency
• streamkit://sessions/{id}/logs — recent session log entries
• streamkit://schema/nodes — full node registry as a resource

MCP Prompts

• debug_pipeline — given a session ID, produces a prompt with pipeline state, node statuses, and recent errors for an LLM to diagnose issues
• design_pipeline — given a natural language description, produces a prompt with available nodes and connection rules to help an LLM generate pipeline YAML

Dependencies

• Phase 1 MCP infrastructure (feat: implement MCP server #359)
• Phase 2 mutation tools (feat(mcp): Phase 2 — graph mutation tools #360) — optional but recommended