feat(bank-connectors): pluggable architecture for bank integrations

[sungdark]

Summary

Implements a pluggable bank connector architecture for FinMind, fulfilling all acceptance criteria from Issue #75:

✅ Acceptance Criteria Met

1. Connector interface — Abstract BankConnector base class (base.py) defining the full contract:

   • get_auth_status() — verify credentials
   • list_accounts() — enumerate linked accounts
   • get_transactions() — fetch transaction history
   • refresh() — incremental refresh (new transactions only)
   • normalize_transactions() — hook for connector-specific normalization

2. Import & refresh support — Full implementation via BankConnectionService:

   • POST /bank-connections/connections/{id}/import — full import with dry_run, date range filtering, per-account or all accounts
   • POST /bank-connections/connections/{id}/refresh — incremental refresh returning only new transactions since last import
   • Deduplication against existing expenses

3. Mock connector included — MockBankConnector (mock.py) for development/testing:

   • Deterministic transaction generation (configurable via seed)
   • default, empty, and error modes for testing all code paths
   • Realistic transaction data (merchants, amounts, types)

Architecture

packages/backend/app/services/bank_connectors/
├── __init__.py       # Public exports, auto-registers MockBankConnector
├── base.py           # BankConnector ABC, Transaction, Account, ConnectorError
├── registry.py       # ConnectorRegistry singleton (pluggable factory)
├── mock.py           # MockBankConnector implementation
└── service.py        # BankConnectionService (orchestration + DB persistence)

packages/backend/app/routes/bank_connections.py  # REST API endpoints
packages/backend/app/models.py                   # BankConnection, BankConnectionAccount, BankImportRun
packages/backend/app/db/schema.sql              # PostgreSQL schema for new tables

Key Design Decisions

• Pluggable factory pattern via ConnectorRegistry — new connectors register with a name and factory function; no code changes needed to core files
• Connector-agnostic normalization — normalize_transactions() delegates to the existing expense_import pipeline so all connectors benefit from existing deduplication, date parsing, etc.
• Encrypted config storage — Per-user connector credentials stored encrypted in config_encrypted TEXT (placeholder implementation; production should use KMS/Vault)
• Import run tracking — BankImportRun table records every import for audit/debugging

Payment

Bounty: $500 — Issue #75

Payment preference: PayPal (or GitHub Sponsors). Please share PayPal email or payment instructions.