Refactor: Introduce services layer to separate business logic from HTTP client

[RadCod3]

Problem

The FireflyClient in clients/firefly.py is currently handling too many responsibilities:

1. ✅ HTTP communication (appropriate)
2. ✅ Request/response serialization (appropriate)
3. ❌ Business logic - aggregations, calculations (should be elsewhere)
4. ❌ Multi-call orchestration - making multiple API calls and combining results (should be elsewhere)
5. 🤔 Model conversion (debatable)

Examples of business logic in the client:

• get_budget_spending(): Aggregates data from multiple limit entries, calculates percentages
• get_budget_summary(): Orchestrates multiple API calls and combines results
• create_bulk_transactions(): Orchestrates multiple individual creates

This violates the Single Responsibility Principle and makes the client harder to test and maintain.

Proposed Solution

Introduce a services layer between the MCP tools and the Firefly client:

MCP Tools → Services → Firefly Client → Firefly API

New Structure

src/lampyrid/
├── tools/              # MCP interface layer (thin)
│   ├── accounts.py     # MCP tool definitions only
│   ├── transactions.py
│   └── budgets.py
├── services/           # NEW: Business logic layer
│   ├── __init__.py
│   ├── accounts.py     # AccountService - orchestration, aggregation
│   ├── transactions.py # TransactionService
│   └── budgets.py      # BudgetService
├── clients/            # API client layer (thin)
│   └── firefly.py      # Pure HTTP client, returns raw Firefly models
├── models/
│   ├── firefly_models.py
│   └── lampyrid_models.py
└── config.py

Responsibility Distribution

Tools Layer (MCP interface):

• Register MCP tools
• Input validation (Pydantic)
• Call service methods
• Return results directly

Services Layer (Business logic):

• Orchestrate multiple client calls
• Business logic (aggregations, calculations)
• Model conversion (Firefly → LamPyrid)
• Return LamPyrid models

Client Layer (HTTP only):

• HTTP communication
• Request parameter formatting
• Response parsing to Firefly models
• Error handling (HTTP status)
• Return raw Firefly models