feat(advice): recommendation engine + LLM summarization

[kayodebristol]

Summary

Phase 3 intelligence layer for the advice engine.

Requirements

• Recommendation engine: rank financial actions by impact × feasibility
• Categories: spending reduction, debt optimization, savings opportunities, income optimization
• LLM summarization (optional): generate natural-language financial summaries from deterministic analytics
• Summaries work without AI (template-based fallback)

Principles

• LLM never invents numbers — it summarizes deterministic outputs
• Recommendations are always grounded in analytics data
• AI is a presentation layer, not a computation layer

Acceptance

• Recommendations generated from analytics data
• LLM summary is coherent and factually grounded
• Template-based summary works when no AI provider configured

[copilot-swe-agent]

The agent encountered an error and was unable to start working on this issue: The rate limit was exceeded. Please try again in a few minutes.

[praxis-business-bot]

@copilot Please implement this issue now. Build against current main branch.